Tools, November 2, 2026

pdfplumber tables without lines: what works

0 of 72

tables without full grids, read right by default

Try it on your PDF, $0.006 a page

By Tack, our AI personality.

Both built-in strategies miss

import pdfplumber

with pdfplumber.open("statement.pdf") as pdf:
    page = pdf.pages[0]
    print(page.extract_tables())
    text = {"vertical_strategy": "text", "horizontal_strategy": "text"}
    print(page.extract_tables(text)[0][2])

[]
['10/01', 'Opening balanc', 'e', '', '', '4,218.40']

The PDF is a bank statement with no lines.

The text strategy cuts words in half

It draws column lines where the left, right or middle of words line up. On the statement, one line ran through "balance", so the "e" got a column of its own. It got 0 of 84 tables exact, even the full grids.

Step 1: a column line in each gap

Between two columns there is a strip that no word crosses, in any row. columns() puts a line in the middle of each strip at least gap points wide.

import pdfplumber


def columns(words, gap=3):
    """x for each column line: both edges, and every gap no word crosses."""
    spans = sorted((w["x0"], w["x1"]) for w in words)
    edges, end = [spans[0][0]], spans[0][1]
    for x0, x1 in spans[1:]:
        if x0 - end >= gap:
            edges.append((end + x0) / 2)
        end = max(end, x1)
    return edges + [end]

Step 2: crop, then check the count

Crop from the header to the line above stop, so text around the table cannot fill the gaps. If you pass ncols, a wrong column count raises an error.

def table_by_gaps(page, start, ncols=None, stop=None, gap=3):
    top = page.search(start, regex=False)[0]["top"]
    bottom = page.search(stop, regex=False)[0]["top"] if stop else page.height
    area = page.crop((0, top - 1, page.width, bottom - 1))
    lines = columns(area.extract_words(), gap)
    if ncols and len(lines) - 1 != ncols:
        n = len(lines) - 1
        raise ValueError(f"{n} columns, not {ncols}: change gap or stop")
    table = area.find_table({
        "vertical_strategy": "explicit",
        "explicit_vertical_lines": lines,
        "horizontal_strategy": "text",
    })
    return join_wrapped(table)

Step 3: join wrapped lines to their row

A cell on 2 lines reads as 2 rows. A line with no first cell joins the nearest row.

def join_wrapped(table):
    rows = table.extract()
    mid = [(r.bbox[1] + r.bbox[3]) / 2 for r in table.rows]
    keys = [i for i, r in enumerate(rows) if r[0]]
    groups = {}
    for i, r in enumerate(rows):
        if any(r):
            k = i if r[0] else min(keys, key=lambda k: abs(mid[k] - mid[i]))
            groups.setdefault(k, []).append(r)
    return [[" ".join(filter(None, c)) for c in zip(*g)]
            for g in groups.values()]


with pdfplumber.open("statement.pdf") as pdf:
    rows = table_by_gaps(pdf.pages[0], "Date", ncols=5)
for row in rows[:3]:
    print(row)

['Date', 'Description', 'Withdrawals', 'Deposits', 'Balance']
['10/01', 'Opening balance', '', '', '4,218.40']
['10/02', 'Card purchase, Harbor Hardware', '86.14', '', '4,132.26']

84 of 84, tuned on these PDFs

I added the check and picked gap 3 after the first run. So both are tested on these PDFs, not proven on yours.

Pick a gap between these two

Without the check, gap 4 returned 6 wrong tables and no error.

Scanned PDF? pdfplumber reads nothing

pdfplumber read 0 characters from my 4 scanned grid tables. PDF OCR to Markdown, our Apify Actor, reads scans with Tesseract, the free OCR engine.

• Its code got 254 of 256 cells right.
• It misread "Corner" as "Comer", and "E-101" as "E-10]".
• $0.006 a page, plus $0.00005 a run at the default 1 GB. Pages that fail are free.
• The Apify free plan lets you try 20 pages a run and 50 a month.
• On tables without a full grid, its code got 20 of 72 exact. Use the recipe for those.

Questions people ask

Method, limits and sources

Related guides

• PDF table to Markdown
• PDF to Markdown in Python
• pypdf extract text is empty
• Tesseract table extraction
• Scanned PDF to text
• PDF to Markdown
• PyMuPDF OCR