Free Online PDF Table Viewer
Drop a PDF below and the tables inside it are found and listed. Pick one and it opens in a real grid: sort, filter, column statistics, copy a range, export as CSV or Excel. The detection runs in your own tab, so the statement never leaves your machine.
A PDF does not contain a table
This is the thing worth understanding before you extract anything from one. A PDF is a description of marks on a page. It says: put the glyph "4" at these coordinates in this font at this size, then put a "2" next to it, then draw a line from here to here. There is no concept of a table, a row, a column or a cell anywhere in the file. What you see as a table is an arrangement of characters that a human eye assembles into one.
Which is why copying a table out of a PDF and pasting it anywhere produces the mess it does. The clipboard gets characters in the order they were drawn, which is often column by column rather than row by row, and all the structure your eye was supplying is gone.
Recovering a table means reconstructing it from the two things the file does carry. The first is ruling lines: when a document actually draws the grid, those lines give exact cell boundaries and the reconstruction is close to certain. The second is alignment: when there are no lines, a column reveals itself because a run of text on successive rows starts at the same horizontal position. Both signals are used here, and ruled boundaries are preferred where they exist because inference is only ever inference.
All of that runs in your browser tab. That is not just a privacy nicety for this format: the PDFs people most need tables out of are bank statements, invoices, payslips, medical results and legal filings, which is the exact category that should not be uploaded to a free web service in exchange for a spreadsheet.
What is on the page
Every table found, listed
Each detected table appears in the picker with its row and column count, and a warning says how many were found. A twelve-month statement gives you twelve entries rather than one and a shrug.
Ruled boundaries preferred
Where the document draws its grid, those lines are used, because a drawn boundary is a fact and an alignment is a guess.
Alignment as the fallback
Unruled tables are recovered from where the text sits, which handles the majority of statements and reports that use whitespace rather than lines.
A real grid, not a text dump
Sortable headers, filter as you type, and a row counter. Sorting a statement by amount to find the largest transaction takes one click.
Column statistics
The sigma in a header gives blanks, distinct count, min, max, mean, median and sum, which is how you check a statement's total against its own line items.
Values kept as printed
Account numbers with leading zeros, reference codes and amounts with currency symbols all come through exactly as the page showed them.
Freeze and select
Pin the first columns so the date stays visible, and drag to select any rectangle of cells for copying.
Copy or export
Copy as TSV, CSV, Markdown or JSON for a selection; Download CSV or .xlsx for the current view, quoted correctly.
Password-protected files
A PDF that asks for a password prompts for one in the page. The password is used to decrypt locally and is never sent anywhere.
A worked example
The Try with sample data link loads a report PDF with tables in it. The first thing you get is not the grid but the count: a message saying how many tables were detected, and a picker listing them. That is the number worth reading first, because a document you thought had one table often has three, and the one you wanted may not be the one that opened.
Pick the table you want and it draws. Now do the check that matters, which no extraction tool can do for you: look at the grid, look at the page, and see whether the columns line up. A well-ruled table will be exact. An unruled one is a reconstruction, and reconstructions go wrong in a specific way, which is that one column splits into two or two merge into one where a value was unusually wide.
Once you trust it, the grid earns its keep. Click the sigma on the amount column and you get the sum of the whole table, which you can compare against the total printed on the page. Sort descending to find the largest line. Filter to a description to see how often something recurs. These are the questions people open bank statements for, and they normally require getting the data into a spreadsheet first.
Then Download CSV, which writes the current view with proper quoting, so a description containing a comma arrives as one cell. If a column did come out split, the CSV editor is the quickest place to repair it before you pass the file on.
Where PDF extraction goes wrong
- Scans. A scanned page is an image, so it goes through optical character recognition in your browser, and a PNG or JPG of a table dropped here does too. Recognition is slower and less reliable than a text layer; check the result against the original.
- Cells that wrap onto two lines. A long description that wraps looks like two rows to a detector reading vertical position. Ruled tables survive this; unruled ones often do not.
- Tables that span pages. A table continuing across a page break, with the header repeated, usually detects as two tables. Export both and concatenate them.
- Multi-column page layouts. A report laid out in two columns of prose with a table inside one of them can confuse the column detection, because the page has real columns that are not table columns.
- Right-aligned numbers next to left-aligned text. Mixed alignment inside one column is the single most common cause of a column splitting in two, because the detector is reading exactly the alignment that the document is being inconsistent about.
None of this is a reason not to extract. It is a reason to look at what came out before you rely on it, which takes about ten seconds and is the difference between a correct spreadsheet and a confidently wrong one.
Frequently Asked Questions
How do I get a table out of a PDF?
Drop the PDF on the box at the top of this page. Every table found in the document is listed in a picker, the first opens in a grid, and you can sort it, filter it, read column statistics and export it as CSV or Excel. The detection runs in your own browser tab, so nothing is uploaded and no account is needed.
How does it find the tables?
A PDF does not contain tables, only characters placed at coordinates, so the structure has to be reconstructed. Two signals are used: ruling lines, which give exact cell boundaries when the document draws them, and the alignment of text itself, which reveals columns when it does not. The detector prefers ruled boundaries where they exist because they are unambiguous.
Does it work on scanned PDFs?
A scan is an image of a page, so there are no characters to read directly. The viewer runs optical character recognition in your browser on such pages, and reads a PNG or JPG of a table the same way. Recognition is slower and less reliable than a real text layer, so a table recovered that way should always be checked against the original.
What if the document has several tables?
All of them are listed with their row and column counts where those are known, and a warning above the grid says how many were detected so you are never left assuming there was only one. Switching between them redraws the grid. A statement with one table per month gives you one entry per month.
Are numbers and dates preserved exactly?
Yes. Every cell shows the exact text the PDF contained, so an account number with leading zeros or an amount written with a currency symbol comes through as written. The type badge on each header is a label and a sort order, not a conversion, so nothing is quietly renumbered on the way in.
Why is one of my columns split in two?
Usually because a value in that column is unusually wide or unusually narrow and shifts the alignment the detector was reading. Unruled tables are inference, not certainty. Compare the grid against the page before you trust an export, and if a table is badly recovered the CSV editor is the quickest place to merge two columns back together.
Can I export what I see?
Yes. Download CSV and .xlsx write the current view, so sorting and filtering first gives you exactly the rows you want rather than the whole table. For a fragment, select a range of cells and use Copy as for TSV, CSV, a Markdown table or JSON.
Is my PDF uploaded?
No. There is no upload endpoint on this page. The document is parsed by JavaScript in your own tab and nothing is transmitted or stored. Bank statements, invoices, medical results and legal filings are the PDFs people most often need tables out of, and they are exactly the ones that should not be handed to a web service.
Related tools and guides
Get the table out of the PDF
Every table found and counted, a real grid to check it in, and nothing uploaded anywhere.
Back to the viewer