How to Extract Tables from a PDF

Why PDF tables are so hard to extract

A PDF has no concept of a table. What looks like rows and columns is just individual text fragments positioned at x/y coordinates, sometimes with drawn lines between them. The "table" exists only in your eyes — the file stores no cells, no rows, no header.

Every extraction method is therefore reconstruction: an algorithm guessing which text fragments belong to the same cell, which cells form a row, and where columns begin. Ruled tables (with visible grid lines) are easier to reconstruct; whitespace-separated tables, merged cells, multi-line cells, and multi-page tables are where most tools fail.

Scanned PDFs add another layer: the page is a photograph, so there is no text at all until OCR runs. Pick your method based on which of these problems your document has.

Method 1: Copy-paste into Excel (quick, small ruled tables)

For a simple, well-ruled table, select it in your PDF viewer, copy, and paste into Excel or Google Sheets. If everything lands in one column, use Data → Text to Columns (Excel) or Data → Split text to columns (Sheets) to re-split the values.

A useful variation: paste into a plain-text editor first to see what structure survived. If the values are separated by consistent spaces or tabs, Text to Columns will recover the table; if the rows are interleaved or truncated, no amount of splitting will fix it.

When this works: single-page ruled tables with short cell values. When it fails: merged cells, multi-line text inside cells, numbers with thousands separators being split apart, and any scanned document.

Method 2: Excel Power Query or Word (built-in, no extra tools)

Modern Excel can import PDF tables directly: Data → Get Data → From File → From PDF. Excel scans the document, lists every table it detects, and shows a preview — pick the table, click Load, and it lands as a proper spreadsheet range. This is the best built-in option on Windows (it requires a Microsoft 365 or 2021+ license and is not available in Excel for Mac in all versions).

Word can also open PDFs directly (File → Open → select the PDF). It converts the document to an editable file, and simple tables usually survive as real Word tables you can copy into Excel.

When this works: digitally-created PDFs with clearly ruled tables. When it fails: whitespace-only tables are often missed or mis-split, multi-page tables import as separate fragments, and scanned PDFs produce nothing — Power Query does not OCR.

Method 3: Python libraries — camelot, tabula-py, pdfplumber (developers)

If you need to extract tables from many PDFs programmatically, the established open-source options are camelot (best on ruled tables, two detection modes), tabula-py (a wrapper around the Java Tabula engine, solid on consistent layouts), and pdfplumber (lower-level control over words, lines, and rectangles — best when you need custom logic).

A minimal camelot example: pip install camelot-py[cv], then tables = camelot.read_pdf("report.pdf", pages="all") and tables[0].df gives you a pandas DataFrame. Export with .to_csv() or .to_excel().

The catch: each library has a layout style it favors, and real-world documents mix styles. Expect to tune settings per document family (flavor="stream" vs "lattice" in camelot, area hints in tabula), and none of them handle scanned PDFs — you would need to run OCR first and lose the layout in the process.

For a broader look at PDF parsing in Python beyond tables, see our guide on converting PDF to Markdown in Python.

Method 4: AI-powered extraction (scanned PDFs, mixed layouts, scale)

The newest approach uses document-understanding models that read the page the way a person does — detecting table structure from the visual layout rather than relying on ruling lines or coordinate heuristics. This is the only method class that handles scanned documents, photographed pages, and inconsistent layouts in one pass.

ParseJet works this way: upload a PDF at parsejet.com/tools/pdf-to-markdown and tables come back as clean Markdown tables in reading order — scanned pages are OCRed automatically. If you want data instead of text, the API returns the parsed document as structured output (see the PDF to JSON tool), and one POST request per file makes it practical for batch pipelines.

When this works: scanned and photographed documents, mixed ruled/whitespace layouts, and batch processing where per-document tuning is impossible. The trade-off: it is a hosted service — for a one-off table on a clean PDF, Method 1 or 2 is faster.

Which method should you use?

One small ruled table, one time: copy-paste (Method 1), falling back to Excel Power Query (Method 2) if the paste jumbles.

Recurring reports with the same layout: Power Query for spreadsheet users, camelot or tabula-py (Method 3) for developers — a one-time setup pays off on every future file.

Scanned documents, photos, or messy mixed layouts: AI extraction (Method 4) is realistically the only option that does not involve retyping.

Hundreds of documents with varying layouts: Method 4 via API — per-document tuning of Method 3 libraries does not scale past a handful of layout families.