How accurate is this?

It depends entirely on the PDF, and it is worth understanding why. A PDF contains no table — it contains characters at coordinates. The grid you see is drawn lines, and nothing in the file records which text belongs to which column. So the columns are inferred from alignment: where text starts, across the whole page. On a real table — an invoice, a bank statement, an exported report — that recovers the structure well, because the producer aligned the columns. On prose or a heavily designed layout, it will not.

My PDF came back as one column.

That means no consistent column alignment was found, which almost always means the page is prose rather than a table. The text is all there in one column — nothing was lost — but there was no grid to recover. If you only want the words, PDF to Text or PDF to Word will give you a better-shaped result.

Do numbers come out as numbers?

Yes. Values that look numeric are written as real numbers so you can sum and sort them immediately, including ones carrying thousands separators, a leading currency symbol, or a trailing percent sign. Everything else is written as text.

My PDF is a scan. Will this work?

No. A scan is a photograph of a page with no text in it at all, so there is nothing to read and no columns to detect. Extracting it would need OCR, which this tool deliberately does not do. The tool detects this case and tells you rather than handing back an empty workbook.

Should I choose one sheet per page or everything in one?

One sheet per page keeps each page's table separate and is the safer default. Choose a single sheet when one long table runs across many pages — a statement or a ledger — so the rows end up contiguous and sortable.

No. The text layer is read with pdf.js in your browser and the .xlsx is zipped together in memory. Statements, invoices and reports stay on your machine.

PDF to Excel — Extract PDF Tables to XLSX

There is no table in a PDF

That sentence is the whole tool. When you look at a table in a PDF you see rows, columns and borders. What the file actually contains is a set of characters, each with a position, plus some lines drawn on the page. There is no record anywhere that says "these five values are a row" or "this is the Amount column".

Every PDF-to-Excel converter, including the expensive ones, is therefore guessing. What separates a good result from a bad one is what the guess is based on.

How the columns are found

The strongest available evidence is alignment. Whoever produced the document lined its columns up, and that alignment survives into the coordinates.

So every text run's left edge is collected across the page and those positions are clustered: runs starting at roughly the same place belong to the same column. A cluster has to be used by more than one row before it counts, which stops a single indented line inventing a column and shifting everything after it. Rows come from shared baselines, and each run is then placed in the column it starts at.

The tolerances scale with the text size, so a six-point bank statement and a fourteen-point report both group sensibly rather than needing a setting you would have to guess at.

Where it works well, and where it does not

Well: invoices, bank and card statements, exported reports, price lists, timetables, results tables — anything generated by software from data, which is most tables that end up in a PDF.

Poorly: prose, which correctly comes out as a single column; multi-column magazine layouts; and tables where cells span columns, since the merge only ever existed in the original document.

A cell whose text wrapped onto two lines in the PDF becomes two rows, because on the page that is genuinely what it is. Expect a little tidying after opening the file — the point is to skip the retyping, not to skip reading the result.

Numbers stay numbers

A workbook where every figure is text is nearly useless: nothing sums, nothing sorts. Values that look numeric are written as real numbers, including ones with thousands separators, a leading currency symbol or a trailing percent sign. That means you can select a column and get a total straight away.

Not OCR

This reads the text layer that is already in the file. A scanned page has none — it is a picture — and no amount of column detection helps when there are no characters to position. The tool reports that clearly instead of producing an empty workbook.

Common uses

Getting a bank or card statement into a spreadsheet for reconciliation
Turning an invoice's line items into rows you can total
Recovering data from a report whose source file is gone
Pulling a published results or price table into a working file

Privacy

Everything runs in your browser. Financial documents are the most common input to a tool like this, and they are precisely what should not be uploaded to a stranger's server.

PDF to Excel