How to open a .mat file without MATLAB (5 free ways)

Someone sends you a .mat file: a test log, a Simulink run, the data behind a paper. You don’t have MATLAB, and double-clicking it does nothing useful. This guide compares five free ways to open it, explains why a file opens in one tool and not in another (it is usually the v7.3 format), and shows how to turn the data into a CSV. The quickest is in your browser: the MegaRows viewer opens every version, v7.3 included, with timeseries, tables, timetables and Simulink Datasets. We ran every command and code sample on test files in each MAT-file version. No .mat file of your own yet? Download a sample MATLAB MAT file.

The five ways at a glance

Quick rule: to look at the data and plot it, use a viewer. To process it, use Python or Octave. If the file is full of MATLAB objects other than timeseries, tables and timetables, you need MATLAB itself, and MATLAB Online is the free way in.

Option 1: MegaRows, read it online in your browser

The MegaRows MAT file viewer is a free web page that opens .mat files on your own device, with nothing to install and no account.

Drop the file onto the viewer or click “Choose a file”. The viewer recognises the version, groups the signals that share a time vector into tables and plots the first one against time. You can also open several .mat files at once and plot run A against run B.

Option 2: Python with scipy (v7 and older) or h5py (v7.3)

Python is free and the usual choice when you want to process the data further. The library you need depends on the file version. scipy.io.loadmat reads v4, v6 and v7 files (SciPy’s documentation says “v4 (Level 1.0), v6 and v7 to 7.2”) and stops at a v7.3 file with this error:

NotImplementedError: Please use HDF reader for matlab v7.3 files, e.g. h5py

Install the packages with pip install scipy h5py, then pick the section for your file.

v4, v6 and v7 files: scipy.io.loadmat

from scipy.io import loadmat, whosmat

# List the variables without loading them: (name, size, class)
for name, shape, cls in whosmat("run.mat"):
    print(name, shape, cls)

# Load the file; simplify_cells turns structs into dicts and cell arrays into lists
data = loadmat("run.mat", simplify_cells=True)
simout = data["simout"]
print(list(simout))                      # the struct's fields
t = simout["time"]                       # a NumPy array
sig = simout["signals"][0]               # a struct array becomes a list of dicts
print(sig["label"], sig["values"].shape)

Output for our test file, a Simulink-style log in “Structure with time” format:

simout (1, 1) struct
yout (1, 1) struct
ArrayFmt (300, 3) double
['time', 'signals', 'blockName']
speed (501,)

Without simplify_cells=True, a struct comes back as a NumPy record array and you need indexing such as data["simout"]["time"][0, 0] to reach a field. To read only some variables from a big file, pass variable_names=["simout"].

v7.3 files: h5py

A v7.3 file is an HDF5 file with a 512-byte MATLAB header in front, so h5py opens it directly. You get MATLAB’s raw storage, though, with three quirks:

• Arrays come back transposed. MATLAB stores arrays column by column, so a 300×3 matrix reads as (3, 300). Add .T.
• Text is stored as numbers: UTF-16 character codes, not strings.
• Cell arrays and struct arrays hold references to data in a hidden #refs# group, which you follow with f[ref].

import h5py

with h5py.File("run_v73.mat", "r") as f:
    print(list(f.keys()))                # variable names (#refs# is MATLAB's own)
    a = f["ArrayFmt"][()]                # read a whole variable into NumPy
    print(a.shape)                       # (3, 300): MATLAB's 300x3, transposed
    a = a.T

    t = f["simout/time"][()].ravel()     # struct fields are paths
    name = f["simout/blockName"][()]     # text is stored as UTF-16 codes
    print("".join(chr(c) for c in name.ravel()))

    ref = f["simout/signals/values"][0, 0]   # struct arrays and cells hold references
    print(f[ref][()].T.shape)

Output for the same test log in v7.3 format:

['#refs#', 'ArrayFmt', 'simout', 'yout']
(3, 300)
model/To Workspace
(501, 1)

If you would rather get dictionaries like loadmat’s, the mat73 package (pip install mat73) does the work for you: mat73.loadmat("run_v73.mat") returned our test file with the arrays the right way round, text decoded and references followed.

Option 3: GNU Octave

Octave is a free, open-source program whose language is largely compatible with MATLAB’s, and its load command reads MAT-files directly. Install it from octave.org (Windows, macOS and Linux), then:

whos -file run.mat                     % list the variables without loading them
s = load("run.mat");                   % load everything into a struct
s.simout.signals(1).label              % reach into structs as in MATLAB
plot(s.simout.time, s.simout.signals(1).values)
csvwrite("speed.csv", [s.simout.time, s.simout.signals(1).values])

In our tests with Octave 11.1, v4, v6 and v7 files loaded fine, structs and cell arrays included. v7.3 is another story: Octave’s manual says it “does not yet implement MATLAB’s v7.3 binary data format”, and with our v7.3 test files it loaded only plain numeric matrices, skipped struct and cell arrays, returned text as numbers, and gave up on one file with an error.

Option 4: MATLAB Online

MATLAB Online runs the full MATLAB in your browser, and its basic version is free with a MathWorks account: 20 hours per calendar month, sessions that end after 15 idle minutes, and 5 GB of MATLAB Drive storage. Upload the file to MATLAB Drive at matlab.mathworks.com, then double-click it or type load. Every version opens, and so do tables, timetables and Simulink objects.

The catch is the upload: your file is copied to MathWorks’ cloud. Check your company’s rules before you upload test or customer data. Once the 20 hours are used up, access stops until the first day of the next month.

Option 5: Simulink Data Explorer for VS Code

In August 2026 MathWorks released Simulink Data Explorer, a free, open-source extension for Visual Studio Code that reads Simulink models, data dictionaries, projects and MAT-files with “no MATLAB or Simulink installation required”. Install it from the Extensions view or a terminal:

code --install-extension mathworks.simulink-data-explorer

A .mat file then opens read-only as a tree-structured table of its variables, and a matrix such as <2x3x2 double> opens in a spreadsheet-style grid. Only files saved with -v6 or -v7 are supported (v7.3 files are reported as unsupported), and there are no charts.

Other options

• HDFView. The HDF Group’s free desktop viewer opens v7.3 files, because they are HDF5 files. You see the same raw layout as in h5py: transposed arrays, text as numbers and references.
• R. The R.matlab package’s readMat() reads v4 to v7 files, compressed ones included, but not v7.3. For v7.3, use an HDF5 package such as rhdf5.
• Julia. MAT.jl reads every version from v4 to v7.3 with matread("file.mat").
• Online viewers and converters. No install. Some send your file to their server and some read it in your browser, and size limits differ: read their privacy wording and watch the Network tab of your browser’s developer tools before you drop work data into one.

Which version is my .mat file?

Most problems come down to the version. Every MAT-file saved as v6 or later starts with a 128-byte header whose first 116 bytes are text. On macOS or Linux, print it in a terminal:

head -c 128 run.mat

A few binary bytes follow the text and may show as odd characters. Compare the start with these two examples:

MATLAB 5.0 MAT-file, Platform: GLNXA64, Created on: Sun Oct  4 12:00:00 2026

MATLAB 7.3 MAT-file, Platform: GLNXA64, Created on: Sun Oct  4 12:00:00 2026 HDF5 schema 1.00 .

• “MATLAB 5.0 MAT-file” means v6 or v7. Both use MATLAB’s Level 5 format (v7 adds compression), and every tool in this guide reads them.
• “MATLAB 7.3 MAT-file … HDF5 schema” means v7.3. Use MegaRows, h5py or mat73, MATLAB Online or another HDF5 tool.
• No readable text means MATLAB’s oldest format, v4 (which scipy, Octave and MegaRows read), or a file that isn’t from MATLAB: a few unrelated programs use the .mat extension too.

On Windows, or anywhere with Python, print the same text with one line:

python -c "print(open('run.mat', 'rb').read(116).decode('latin-1'))"

or let SciPy decide:

from scipy.io.matlab import matfile_version

print(matfile_version("run.mat"))   # (0, 0) = v4, (1, 0) = v6 or v7, (2, 0) = v7.3

Why v7.3 needs HDF5 tools. Since MATLAB R2006b, save -v7.3 writes an HDF5 file with MATLAB’s 512-byte header in front; the HDF5 signature sits at byte 512. MATLAB needs this format for variables of 2 GB or more, and its matfile function creates v7.3 files by default, while a plain save still writes v7. Level 5 readers can’t parse HDF5, and HDF5 libraries see MATLAB’s raw storage, which is why scipy.io.loadmat, R.matlab and the VS Code extension can’t read v7.3 files and h5py shows their arrays transposed. For a general .h5 or NetCDF-4 file, use the HDF5 dataset viewer; use the MAT viewer for MATLAB structures.

Tables, timetables and Simulink objects

Some variables are hard to open anywhere but MATLAB. Tables, timetables, timeseries, datetime, string and categorical arrays, and Simulink’s Dataset and SimulationOutput objects are stored in an undocumented, MATLAB-only form. SciPy returns them as opaque placeholders and h5py shows a short array of numbers tagged with the class name. MegaRows reads timeseries (alone, in structs, in a tscollection or in a logged Dataset), whole simulation outputs, Simulation Data Inspector exports, tables, timetables, and datetime, duration, string and categorical arrays, from v7 and v7.3 files alike; it lists the other objects (such as containers.Map and your own classes) in the notes and opens everything else in the file.

This matters for Simulink logs: the To Workspace block saves in Timeseries format by default, and new models return everything in a single SimulationOutput object called out. MegaRows opens out saved as it is, with its logsout, yout, xout, tout and To Workspace variables each in their own tables. For the other tools here, ask whoever made the file to run these MATLAB lines: they produce a file that every tool can read.

% A table becomes a struct with one field per column
s = table2struct(T, 'ToScalar', true);
save('data.mat', '-struct', 's', '-v7')

% Simulink: set the To Workspace block's Save format to "Structure With Time", then
simout = out.simout;
save('run.mat', 'simout', '-v7')

For a timetable, call timetable2table first.

Convert a .mat file to CSV or Excel

Excel can’t open a .mat file, but it opens CSV. There are two easy ways to get one.

With MegaRows

Use the free tool to convert MAT to CSV: open the file, tick the signals you need (up to eight per chart) and choose Do more › Download chart data (CSV). You get the plotted signals, time first, for the range in view: zoom out fully for the whole recording, or zoom in for a single event. There is no direct Excel export yet, and Excel stops at 1,048,576 rows per sheet; for longer recordings, see how to open a CSV file that’s too big for Excel.

With Python and pandas

This turns the matrix and channel names in our sample Simulink-style log into a CSV and an Excel workbook (install pandas and openpyxl first):

import pandas as pd
from scipy.io import loadmat

data = loadmat("megarows-sample-simulink.mat", simplify_cells=True)
df = pd.DataFrame(data["simout"], columns=data["simout_names"])
df.to_csv("simout.csv", index=False)
df.to_excel("simout.xlsx", index=False)    # needs openpyxl
print(df.shape)                            # (12001, 7)

For a v7.3 file, read the arrays with h5py or mat73 as shown above, transpose them, and pass them to pd.DataFrame the same way. In Octave, csvwrite does the same job.

Going the other way, from a CSV to a .mat file? See how to convert CSV to a MATLAB .mat file.