PDF Table Extractor

How to use the pdf table extractor

1. Choose a PDF containing a table.
2. Check the detected rows, and adjust the column gap if the split is wrong.
3. Copy or download the table as CSV.

About this tool

A PDF contains no tables. It contains glyphs at coordinates, and the ruled lines a reader sees are drawn separately from the text they appear to contain — there is no relationship in the file between the two. Extracting a table therefore means inferring one from where the words sit, which is a guess, and a well-behaved tool says so.

The inference is straightforward: text sharing a baseline is a row, and a horizontal gap wider than a threshold starts a new column. Both thresholds are adjustable, because a densely set table needs a smaller gap than a spacious one, and the same setting cannot suit both.

What makes the result checkable is the grid-filled percentage. A real table fills most of its cells; prose that happens to have wide word spacing produces a sparse, ragged grid, and a low percentage is the signal that the detection has found a table where there is not one. It is reported rather than hidden, and anything under sixty per cent is flagged.

Words broken mid-token for kerning are rejoined, since PDF text is frequently emitted in fragments that have nothing to do with word boundaries. A scan has no text layer at all, so nothing can be recovered from one without OCR, and the page says that rather than returning an empty result.

Common uses

• Getting a price list or a statement out of a PDF and into a spreadsheet.
• Recovering tabular data when only the PDF survives.
• Checking whether a document's tables can be extracted at all.

Frequently asked questions

Why is the extraction sometimes wrong?

Because a PDF has no table structure to read. Rows and columns are inferred from text positions, and closely spaced columns or ragged layouts can be split incorrectly.

What does the grid-filled percentage mean?

How much of the detected grid actually has content. A real table fills most of its cells; a low figure usually means prose with wide spacing was mistaken for a table.

Can it read a scanned table?

No. A scan is an image with no text layer, so there are no positions to work from. That needs OCR first.