Home / Guides
File rescue

Open an ACCDB or MDB File Without Microsoft Access

By · November 11, 2026 · 13 min read

The short version

An ACCDB or MDB file is a Microsoft Access database: tables, queries and forms in one file. To open an ACCDB file without Access, on Windows use Excel's Data, Get Data, From Database, From Microsoft Access Database, which Microsoft documents as the way to pull a table or query into a sheet. On a Mac or Linux, install the free mdbtools and run mdb-tables to list the tables and mdb-export to get each one as CSV. If you only want the data in a spreadsheet, the free Access to Excel converter on this site turns an .accdb or .mdb into one .xlsx per table. Watch the dates: mdbtools on its own prints dates before 1900 and far-future dates as 1900-01-00, which is not a date.

What is an ACCDB file, and what is an MDB file?

Both are Access databases. .mdb is the older format. Microsoft's page Which Access file format should I use? says that starting with Access 2007, .accdb is the default, and that .mdb was used "before the .accdb file format was introduced in Access 2007." The same page lists what .accdb added: multivalued fields, the Attachment data type, and from Access 2010 a Calculated data type. It also lists what .accdb gave up: replication and user-level security, which only .mdb supports.

The engine underneath changed name with the format. Microsoft's SQL Server import wizard page for Access lists the providers side by side: "Microsoft Jet Database Engine" for files from Office versions earlier than 2007, and "Microsoft Access Database Engine" (ACE) for Office 2007 and later, up to Microsoft.ACE.OLEDB.16.0 for Office 2016.

There is more than one .mdb. mdb-ver from mdbtools reads the version out of the file header, and its manual page maps the answers: JET3 for Access 97, JET4 for Access 2000, XP and 2003, and ACE12 to ACE16 for Access 2007 to 2016. I ran it on three test databases from the mdbtools project's public test data repository:

$ mdb-ver --version
mdbtools v1.0.1
$ mdb-ver DateTestDatabase.mdb
JET4
$ mdb-ver nwind.mdb
JET3
$ mdb-ver ASampleDatabase.accdb
ACE12

The extension tells you the family, mdb-ver tells you the generation. ACCDB vs MDB: which Access format you have goes further into telling them apart and why it matters.

How to open an ACCDB file on Windows without Access

Two Microsoft routes, neither of which needs a copy of Access. I have no Windows machine for this article, so these are Microsoft's documented steps, not something I ran.

Excel. The Power Query page Import data from data sources gives the path: Data, Get Data, From Database, From Microsoft Access Database. Pick the .accdb, and the Navigator shows the tables and queries; select one to preview it, then Load or Transform. That is the "accdb to excel" answer for anyone with Excel on Windows, and saving the sheet as CSV from there is the "accdb to csv" answer.

The Access Database Engine. If Excel or another program says it cannot find a provider for the file, the missing piece is the engine itself. The import wizard page points to the Microsoft Access Database Engine 2016 Redistributable. It says "the latest version of the components can open files created by earlier versions of Access", so one install covers .mdb and .accdb, and that a machine with 32-bit Office needs the 32-bit engine. On a Microsoft 365 machine the installer may refuse to sit next to Office's click-to-run components; the page's workaround is to run the downloaded AccessDatabaseEngine.exe with the /quiet switch.

There is also the free Access Runtime. Microsoft's runtime download page describes it as the way to distribute Access applications to people without Office, with "all design-related UI" removed or disabled. That suits someone who was handed a finished Access application with forms. For pulling tables out of a file, Excel is the shorter path.

How to open an MDB or ACCDB file on a Mac or Linux

Microsoft's page on ODBC drivers for Excel for Mac only covers SQL Server out of the box; for other sources it points to third-party drivers and does not mention Access. The free route is mdbtools, which its README describes as "a set of programs to help you extract data from Microsoft Access files". It installs with brew install mdbtools on a Mac and apt install mdbtools on Debian or Ubuntu. Everything below is mdbtools 1.0.1 from Homebrew.

List the tables, then look at their columns:

$ mdb-tables -1 ASampleDatabase.accdb
Asset Items
$ mdb-schema -T "Asset Items" ASampleDatabase.accdb
...
CREATE TABLE [Asset Items]
 (
    [Asset No]          Text (5), 
    [Asset Category]            Text (20), 
    ...
    [Acquired]          DateTime, 
    [Cost]          Currency, 
    [Warranty]          Long Integer, 
    ...
    [Comments]          Memo/Hyperlink (255)
);

Then export a table to CSV. mdb-export writes one table to standard output:

$ mdb-export ASampleDatabase.accdb "Asset Items" > asset_items.csv
$ head -3 asset_items.csv
Asset No,Asset Category,Make,Model,Description,Owner,Serial No,Acquired,Cost,Warranty,Tax Scale,Supplier No,Comments
"30050","Computer Hardware","GEO Rocket","220ZX","Computer","Sales","344-667","09/02/97 00:00:00",1995.5000,12,"A","44577",
"30051","Computer Hardware","GEO Blast","Surger 350","Computer","Sales","345-556","09/02/97 00:00:00",2450.0000,12,"A","44577",

Text is quoted, numbers are not, Currency keeps its four decimals. The date is the problem: 09/02/97 could be September or February, and 1997 or 2097 to anything that has to guess. The mdb-export manual has -D and -T to set date and date-time formats with strftime patterns. Always pass them:

$ mdb-export -D '%Y-%m-%d' -T '%Y-%m-%d %H:%M:%S' ASampleDatabase.accdb "Asset Items" | sed -n 2p
"30050","Computer Hardware","GEO Rocket","220ZX","Computer","Sales","344-667","1997-09-02 00:00:00",1995.5000,12,"A","44577",

Leave quoting on. -Q drops the quotes, and in the Access 97 Northwind sample an address with a comma in it then splits into two columns:

$ mdb-export -Q nwind.mdb Umsätze | sed -n 3p
10331,BONAP,9,11/16/94 00:00:00,12/28/94 00:00:00,11/21/94 00:00:00,1,10.1900,Bon app',12, rue des Bouchers,Marseille,,13008,France

To get every table, loop over mdb-tables -1. The flags, ODBC and the rest of the toolkit are the subject of Reading .mdb files on Mac and Linux with mdbtools.

Dates that come out wrong

The fixture our converter's tests use, DateTestDatabase.mdb, is built to break date handling. Its rows are labelled with what they hold. Here is the export with ISO formats forced:

$ mdb-export -D '%Y-%m-%d' -T '%Y-%m-%d %H:%M:%S' DateTestDatabase.mdb DateTest
IDCol,DateTimeCol,DateCol,TimeCol,LongTextCol,ShortTextCol
1,"2021-12-02 00:00:00","2021-12-02 00:00:00","2021-12-02 00:00:00","Hello","Hello"
2,"1900-01-00 00:00:00","1900-01-00 00:00:00","1900-01-00 00:00:00","Year 9999","Year 1999"
3,"1900-01-00 00:00:00","1900-01-00 00:00:00","1900-01-01 00:00:00","Pre-1900 years","Pre-1900 years"
4,"1900-01-00 00:00:00","1900-01-00 00:00:00","1900-01-00 00:00:00","Pre-1000 years","Pre-1000 years"
5,"2021-12-02 00:00:00","2021-12-02 00:00:00","2021-12-02 00:00:00","Escaped new line \r\n in field","Escaped new line \r\n in field"
6,"2021-12-02 00:00:00","2021-12-02 00:00:00","2021-12-02 00:00:00","New line
in field","No new line"

1900-01-00 is not a date. To see what the file really holds, I read the raw bytes. Access stores a Date/Time as a double, and Microsoft's article on how Access stores Date/Time data says 0 is December 30, 1899, earlier dates are negative, and valid values run from -657,434 (January 1, 100) to 2,958,465 (December 31, 9999). A short Python scan for each row's ID followed by its three doubles found:

4 (-463126.0, -462760.0, -462395.0) ['0632-01-01 00:00:00', '0633-01-01 00:00:00', '0634-01-01 00:00:00']
3 (-728.0, -363.0, 2.0) ['1898-01-01 00:00:00', '1899-01-01 00:00:00', '1900-01-01 00:00:00']
2 (2958101.0, 2958101.0, 2958101.0) ['9999-01-01 00:00:00', '9999-01-01 00:00:00', '9999-01-01 00:00:00']
1 (44532.0, 44532.0, 44532.0) ['2021-12-02 00:00:00', '2021-12-02 00:00:00', '2021-12-02 00:00:00']

All of these are inside Access's documented range. mdbtools 1.0.1 printed eight of them as 1900-01-00; only 1900-01-01 (stored as 2.0) came through among the odd ones. So in my run, dates before 1900 and in the year 9999 did not survive mdbtools. Real data has these: a birth year typed as 0632, or 9999-12-31 used as "no end date". Count the 1900-01-00 values in any export before you trust it.

Row 6 shows the other trap: a memo with a line break in it spans two physical lines in the CSV. That is valid CSV inside quotes, but the mdb-export manual lists it under BUGS, and a naive line-by-line script will break the row in two. -e escapes it as \r\n instead.

Pictures, OLE objects and attachments

Access 97 Northwind keeps a bitmap in Categories.Picture, declared OLE (255). mdb-export writes those bytes straight into the CSV by default, which makes the file unreadable in a spreadsheet. The -b option takes strip, raw, octal or hex:

$ mdb-export -b strip nwind.mdb Categories | head -2
CategoryID,CategoryName,Description,Picture
1,"Beverages","Soft drinks, coffees, teas, beers, and ales",""
$ mdb-export -b hex nwind.mdb Categories | sed -n 2p | cut -c1-200
1,"Beverages","Soft drinks, coffees, teas, beers, and ales","151C2F00020000000D000E0014002100FFFFFFFF4269746D617020496D616765005061696E742E5069637475726500010500000200000007000000504272757368

The hex starts with an OLE header that names "Bitmap Image" and "Paint.Picture" before the image itself; it is not a bare .bmp.

Attachment fields, the .accdb-era type, are stored differently. mdb-tables -S (show system tables) on DateTestDatabase.mdb lists a hidden table called f_75A72A8786B149369D90EB962DAB8A2C_Data, and its schema is what an attachment looks like from outside Access:

$ mdb-schema -T f_75A72A8786B149369D90EB962DAB8A2C_Data DateTestDatabase.mdb
...
    [FileData]          OLE (255), 
    [FileFlags]         Long Integer, 
    [FileName]          Text (255), 
    [FileTimeStamp]         DateTime, 
    [FileType]          Text (255), 
    [FileURL]           Memo/Hyperlink (255), 
    [MSysResources_Data]            Long Integer
...

Each row holds a file with its name and type, and the last column's name points back to the parent table, MSysResources, which is a system table. So there were no user attachments to test; I am not going to claim more about attachments than that schema shows.

ACCDB to Excel converter, online and free

The Access to Excel converter on this site runs the same mdbtools commands on the server, with ISO dates, a marker that keeps NULL apart from an empty string and -b hex for OLE columns, and writes one sheet per table. Where mdbtools prints 1900-01-00, it reads the stored double instead and uses it only if that second reading agrees with mdbtools on every other date in the table. I pushed the three files through its Python reader and writers (READERS["access"], then WRITERS["xlsx"] and WRITERS["csv"]) and read the first data row of the .xlsx back with openpyxl: column, value, cell type, number format.

xlsx ['Asset Items.xlsx']
csv ['Asset Items.csv']
Asset No '30050' s General
...
Serial No '344-667' s General
Acquired datetime.datetime(1997, 9, 2, 0, 0) d yyyy-mm-dd h:mm:ss
Cost 1995.5 n 0.0000
Warranty 12 n General
Tax Scale 'A' s General
Supplier No '44577' s General
Comments None n General
$ sed -n 2p "Asset Items.csv"
30050,Computer Hardware,GEO Rocket,220ZX,Computer,Sales,344-667,1997-09-02 00:00:00,1995.5000,12,A,44577,

Asset No and Supplier No are Text in Access and stay text cells (s) in Excel, so they are not turned into numbers. Acquired is a real Excel date, Cost is a number shown with four decimals, and the CSV keeps 1995.5000 as the file has it.

DateTestDatabase.mdb came through with every date, including the eight mdbtools printed as 1900-01-00, and a note saying so:

  ACCESS_DATE_DECODED: DateTest: 8 date value(s) in DateCol, DateTimeCol, TimeCol are valid Access dates that mdbtools could not print (it printed '1900-01-00'). They were decoded from the stored value instead.
$ head -5 DateTest.csv
IDCol,DateTimeCol,DateCol,TimeCol,LongTextCol,ShortTextCol
1,2021-12-02 00:00:00,2021-12-02 00:00:00,2021-12-02 00:00:00,Hello,Hello
2,9999-01-01 00:00:00,9999-01-01 00:00:00,9999-01-01 00:00:00,Year 9999,Year 1999
3,1898-01-01 00:00:00,1899-01-01 00:00:00,1900-01-01 00:00:00,Pre-1900 years,Pre-1900 years
4,0632-01-01 00:00:00,0633-01-01 00:00:00,0634-01-01 00:00:00,Pre-1000 years,Pre-1000 years

Those are the same values the raw-byte scan above found. The memo in row 6 keeps its carriage return and line feed (\r\n) in the CSV, byte for byte as the file holds it.

The .xlsx needs one more look. In the sheet XML, the 1898 and 0632 cells hold negative serial numbers (-728, -463126). openpyxl reads them back as the right dates, but Excel's 1900 date system counts from January 1, 1900 (Date systems in Excel), so check those cells in Excel itself before you rely on them. The CSV has them as plain ISO text.

nwind.mdb gave nine sheets and nine CSVs, one per table, with the OLE pictures as hex:

csv ['Categories.csv', 'Customers.csv', 'Employees.csv', 'Order Details.csv', 'Orders.csv', 'Products.csv', 'Shippers.csv', 'Suppliers.csv', 'Umsätze.csv']
$ head -2 Categories.csv | cut -c1-120
CategoryID,CategoryName,Description,Picture
1,Beverages,"Soft drinks, coffees, teas, beers, and ales",151C2F00020000000D000E0014002100FFFFFFFF4269746D617020496D6167

The size of the pictures matters here. Each Categories.Picture is 21,492 hex characters and fits in a cell. Each Employees.Photo is a 21,626-byte bitmap, 43,252 hex characters in the CSV, and an Excel cell holds at most 32,767 characters, so the .xlsx keeps only the start of each. The converter says so:

  XLSX_CELL_TRUNCATED: An Excel cell holds at most 32,767 characters, so longer values were cut (Employees.Photo: 9). The CSV output keeps them whole.

For pictures, take the CSV.

The upload limit is 10 MB. A bigger database, or one you are moving into a server database, belongs with the Access to SQL Server or Access to PostgreSQL guides.

Sources