You can do a steel takeoff from scanned drawings, but the file type changes the tools that work. Vector PDFs carry real, selectable text that software reads directly; scanned (raster) PDFs are just pictures, so member labels must be read optically. The test takes five seconds: if you can select the text, it's vector.
Half the drawing sets that land in a small fabricator's inbox are scans — renovation jobs from paper archives, sets the GC printed and re-scanned to stamp, 1990s originals that were faxed before they were scanned. If your takeoff process only works on clean CAD exports, you're hand-counting the rest. Here's how to handle both.
What's the difference between a vector PDF and a raster PDF?
A vector PDF comes straight out of CAD or BIM software (AutoCAD, Revit, Tekla). It contains actual drawing instructions — lines, arcs, and, critically, text objects. "W12X26" exists in the file as the characters W, 1, 2, X, 2, 6, at a known position on the page. Software can read every label perfectly, at any zoom, because the text is data.
A raster PDF is an image wrapped in a PDF — a grid of pixels, exactly like a photo. Every scan is raster. The "W12X26" on a scanned sheet is just a pattern of dark pixels; nothing in the file knows it's text. Reading it requires optical character recognition (OCR), and OCR quality depends entirely on scan quality.
| Property | Vector PDF | Raster / scanned PDF |
|---|---|---|
| Source | CAD/BIM export | Scanner, plotter copy, fax, photo |
| Text | Real, selectable characters | Pixels — must be OCR'd |
| Zoom | Infinitely sharp | Blurry past scan resolution |
| File size (per sheet) | Often 0.1–2 MB | Often 2–20 MB |
| Takeoff readability | Near-perfect for software | Depends on DPI, skew, contrast |
| Measurement | Exact coordinates | Only as good as the pixels |
Mixed sets are common and sneaky: a vector base set with three scanned addendum sheets stapled in, or a vector file where the architect's sheets are vector but the structural sheets were scanned from the engineer's paper originals. Check per sheet, not per file.
How do I tell which kind of PDF I have?
Three tests, fastest first:
- Try to select text. Open the PDF, drag across a member label. If "W12X26" highlights as text, that sheet is vector (or has been OCR'd already). If you get a rubber-band selection box and nothing highlights, it's raster.
- Zoom to 800%. Vector linework stays razor sharp; raster goes blocky and gray at the edges.
- Search the file. Ctrl+F for a designation you can see on the page, like "W12". If search finds nothing you can plainly read, the text isn't real text.
Do this before you commit to a bid schedule. A 60-sheet raster set is a different amount of work than a 60-sheet vector set, and knowing that on day one is worth an hour of anyone's time. It also belongs on your PDF blueprint checklist alongside sheet scaling and revision checks.
What goes wrong with scanned drawings?
The classic failure modes, roughly in order of how often they ruin an afternoon:
- Low resolution. 150–200 DPI scans were fine for reading on paper-sized prints, but structural text on a 36x24 sheet is small — at 200 DPI, an eighth-inch-tall "W16X31" is about 25 pixels tall and the difference between 3 and 8 is a few pixels. 300 DPI is the working minimum; 400–600 is comfortable.
- Skew. The paper went through the scanner a degree or two crooked. Humans barely notice; text readers and anyone measuring off the sheet notice a lot.
- Rotation. Sheets scanned sideways or upside down, often inconsistently within one set — page 4 is landscape, page 5 is portrait-rotated-90.
- Generational loss. The worst sets are copies of copies: plotted, photocopied, faxed in 1998, scanned in 2011, re-scanned from a binder last month. Each generation fattens strokes and fills in loops until 6, 8, and B converge.
- Speckle and bleed-through. Dirt, hole punches, coffee, highlighter over the exact callout you need, and text from the sheet behind ghosting through thin paper.
- Hand annotations. Field markups and clouded revisions in pen — sometimes the only record of a change — that no OCR reads reliably and no vector file contains.
Every one of these adds misread risk exactly where it hurts: a misread designation changes weight, and a missed label changes count. These are the same two failure classes behind most steel takeoff mistakes, scan or no scan — raster sets just raise the base rate.
What's the best workflow for a takeoff from scanned drawings?
Practical order of operations:
- Ask for the vector originals first. One email to the GC or engineer — "do you have the CAD-exported PDFs?" — solves the entire problem more often than people expect. Nobody scans for fun; scans usually mean somebody printed unnecessarily.
- If it's paper, control the scan. 300+ DPI, grayscale (not bilevel black-and-white, which destroys thin text), full sheet size, straight feed. A rescan costs minutes; a bad scan costs the whole takeoff.
- Deskew and rotate before anything else. Most PDF tools can batch-deskew. Get every sheet upright and square so measurements and reading both work.
- Verify the scale on every scanned sheet. Scans get resized in copying. Find a dimensioned gridline — say 25'-0" between column lines — and check it against the graphic scale before trusting any measured length. Never trust the stated scale on a copy.
- Read the schedules hardest. On bad scans, the member schedule is your recovery mechanism — if the plan callout is illegible but reads "B1", the beam schedule tells you B1 is a W16X31. How that mapping works is covered in how to read a steel member schedule.
- Flag every uncertain read. Keep a running list of "probably W16X31, could be W16X36" items and close them out with an RFI or a schedule cross-check, not a guess. The weight delta on a guessed digit compounds across every occurrence.
Can takeoff software read scanned steel drawings?
Most can't — and this is the quiet catch in a lot of tool demos. Software built around extracting PDF text finds nothing on a raster page, because there's no text to extract. Upload a scanned set to a text-only tool and it confidently reports zero members, or worse, reads only the two vector sheets in a mixed set and presents that as the whole job. If you're evaluating tools, "what happens on a scanned sheet?" is one of the sharpest questions you can ask — we covered the general version in can AI read steel drawings?
SteelFlo handles both. On vector pages it reads the embedded text directly; on scanned or image-based pages it reads the drawing optically, so a faxed-then-scanned 1997 renovation set gets the same treatment as a fresh Revit export. Every label it finds — either way — is drawn as a bounding box on the sheet, so you verify each detection against the drawing rather than trusting a blind list; the count is the number of boxes, nothing inferred. Detections resolve against 4,500+ sections across 6 standards, and with human verification of flagged items, accuracy runs 95–99% — the box-on-the-drawing review step matters more on scans, because it puts every borderline read in front of your eyes instead of burying it in a spreadsheet. On genuinely rough sheets, expect more flagged boxes to review; that's the tool being honest about scan quality, and it's how AI-plus-verification stays accurate where either alone fails — more on that in AI accuracy on steel takeoffs.
One scope note: no software recovers information that isn't in the pixels. If the scan is 100 DPI and the callout is smeared, the fix is a better scan or an RFI — and drawing legibility is ultimately the design team's responsibility under contract documents standards like AISC's Code of Standard Practice (aisc.org).
Quick answers
How do I know if my PDF is scanned or vector? Try selecting text on the sheet. If member labels highlight as text, it's vector; if nothing selects, it's a scan.
What DPI do scanned drawings need for takeoff? 300 DPI minimum for structural sheets; 400–600 for dense sheets or small text. Below 200, expect misreads.
Can I measure lengths on a scanned drawing? Yes, after verifying scale against a known dimension like a gridline spacing — never trust the title-block scale on a copy.
Can SteelFlo read scanned drawings? Yes — it reads vector text directly and reads scanned/image pages optically, and every detection is shown as a verifiable box on the drawing.
What should I do with illegible callouts? Cross-check the member schedule first, then RFI. Never guess a digit — the weight error repeats on every occurrence.
Got a scanned set sitting in your inbox right now? Upload it to SteelFlo and see every detected member boxed on the drawing — free trial, no credit card. For quick weight lookups while you review, keep the free steel calculator and the section database at /shapes open in a tab.