↑↓ to navigate Enter to open Esc to close

PDF to Excel Free: Preview First, No Upload

No "99% accuracy" claim, because table extraction genuinely varies by document. See the detected grid on screen before downloading — if it does not match your table, you know before committing rather than after opening a broken spreadsheet. Best for PDFs with clear, regular tables; not for scanned tables or ones with merged cells.

Pulls a table's structure out of a PDF into a real, editable XLSX file. Nothing is uploaded.

PDF to Excel

Local
Choose a PDF with a table, or drop it here Nothing leaves this tab
Read by PDF.js, fetched when you add a file. Written by SheetJS, fetched only when you click Download.

How a table gets found without being told where it is

A PDF does not store a table as a table — it stores each piece of text at an exact position on the page, with no record of which pieces belong to the same row or column. Finding the grid means reconstructing it from position alone:

StepWhat happens
RowsText at a similar height on the page is grouped into the same row, allowing a small tolerance for natural baseline variation
ColumnsThe starting position of every piece of text across every row is clustered into a set of column positions
CellsEach piece of text is assigned to whichever detected column its position is closest to
PreviewThe resulting grid is shown before anything is written, so you can check it before downloading

This works well when a document was produced with an actual table — a report, an invoice, a price list — where columns line up consistently. It works poorly on a scanned image of a table, or a layout using merged cells and multi-row headers that do not have consistent column positions to detect.

"99% accuracy" is not a number this tool will claim. Every competitor checked advertises a figure like that for table extraction, which does not hold up for a genuinely hard problem — table structure varies too much between documents for a single accuracy number to mean anything. The preview exists so you judge the actual result, not a marketing claim.

Compared with the alternatives

ToolFile goesAccuracy claim
This pageNowhereNone — see the preview instead
SmallpdfTheir serversNot specified, OCR-based for scans
Several tools checkedTheir servers“99% accuracy” advertised
Adobe AcrobatDepends on planNot specified

What extracts cleanly, and what does not

Table typeResultWhy
A simple grid: consistent columns, one line per rowGoodColumn positions are consistent, so clustering finds them reliably
A table with a multi-line headerMixedHeader rows may not align to the same column positions as the data below
Merged cellsPoorA merged cell has no single column position to detect — it may land in the wrong column or split oddly
A scanned table (image, no text layer)Not detectedThis reads the text layer only; run PDF OCR first if the document is a scan

Why the preview matters more than a setting

Row and column detection uses a small tolerance for natural variation — text on the same printed line is rarely at the exact same height to the point, and two genuinely separate columns still need to be told apart from one wide column with extra spacing. Rather than expose those tolerances as settings to tune, the preview shows the actual outcome so you can judge it directly:

What you see in the previewWhat it means
Rows and columns line up cleanlyThe table is a good candidate — download with confidence
Two values merged into one cellThose columns sit close together on the page; the source document's spacing is the limit here, not a setting
An extra blank row or columnUsually a stray mark or page number caught by the detector — harmless, easy to delete once in the spreadsheet
No table shown at allThe page's text has no consistent grid to detect — check whether it is a scanned page first

Common mistakes

Downloading without checking the preview. The grid shown after “Detect tables” is exactly what will be written — if a column looks wrong there, it will be wrong in the spreadsheet too. Checking first costs a glance; fixing a broken spreadsheet after the fact costs longer.
Trying this on a scanned document. A scan is a picture, with no text positions to cluster into a table. Run PDF OCR first to add a text layer, then this tool has something to work from.
Uploading a financial statement just to get its numbers into a spreadsheet. Bank statements, invoices and financial reports are exactly what people extract, and exactly what should not sit on a server whose retention policy you are trusting. Local extraction removes the question.

Frequently asked questions

Can I adjust how strict the row or column detection is?

There is no setting to tune — the preview exists instead, so you can see the actual outcome directly rather than guessing at a tolerance value. If two columns merge because they sit close together on the page, that is a property of the source document's layout, not something a setting here would fix.

Is my file uploaded?

No. It is read and the spreadsheet is written entirely in your browser — the network panel in developer tools shows no request carrying your document. The only network requests are for the two libraries, and they happen once.

Why didn't it detect my table?

Most commonly because the page is a scanned image rather than real text — this reads the text layer only. Run PDF OCR first if that is the case. It can also happen with layouts that do not have consistent column positions, such as heavily merged cells.

Will it work on any PDF table?

It works best on clear, regular tables with consistent columns — reports, invoices, price lists. Complex layouts with merged cells or multi-row headers may come out wrong. The preview shows the actual result before you download, so you can check rather than guess.

Can I choose which pages to convert?

Yes — type a range like 1-5, 8, or leave it as all. Each page with a detected table becomes its own sheet in the workbook.

Why does one column contain text that should be split into two?

Column detection groups nearby starting positions together; two genuinely separate columns that sit very close together on the page can be merged into one. This is visible in the preview before downloading, so you can catch it there.

Does formatting like bold or colour carry over?

No — this extracts the text and its position, not visual styling. The output is plain values in a grid, ready for formulas and sorting.

Is there a page or size limit?

No cap is imposed. Nothing runs on a server, so there is nothing to ration.

Related tools

See the full list of Document Tools, or try: