What survives when you convert an HWP file
Every converter promises accuracy. The useful question is narrower: which parts of a document survive, which parts drift, and how often does something go wrong badly enough to matter? We measured this against real Korean government documents rather than guessing, and the answers are uneven enough to be worth knowing before you rely on the output.
The short answer
- Text — converts reliably. Korean is preserved as Unicode, so characters are never the problem.
- Tables — structure survives well in both formats. Exact column widths may shift.
- Headings and lists — survive as structure, which is what matters for reading and for machine processing.
- Images — extracted and embedded, at the cost of output size.
- Exact page layout — reproduced for HWPX in most cases, approximated for older binary .hwp, and occasionally wrong in a way worth watching for.
Why layout is the hard part
Text extraction and layout reproduction are different problems, and only the second one is genuinely difficult.
HWPX files saved by Hancom Office embed a typesetting cache — the line and page layout the application had already computed. A converter that reads that cache can place text where the original placed it rather than re-flowing and hoping. This is why an HWPX preview often looks nearly identical to the source document.
Older binary .hwp files generally do not carry that information in a usable form. Conversion yields accurate text, headings, and tables, but the visual page is approximated. If exact appearance matters, ask the sender for a PDF — see sending HWP files to people abroad for what to request.
What we measured
We audited 82 real Korean government HWPX documents — press releases and public notices, the kind of file people actually receive — and checked the layout rendering against the originals.
Roughly 17% rendered with content clipped outside the page boundary. In the worst case a nine-page document rendered as three, losing six pages of content. The text extraction for those same documents was fine; the failure is specific to reproducing visual layout, and it originates upstream in the rendering engine rather than in anything we could fix directly.
That number is why our viewer behaves the way it does. We analyse the rendered coordinates, detect when elements fall outside the page, and then warn you instead of opening the layout view by default. PDF export falls back to a text-based path when clipping is detected. A page that looks faithful but is missing content is worse than a plain one that has everything, because you have no way to know what is absent.
File size, and why the limit is where it is
We measured the size distribution across 98 real documents: median 0.34MB, 95th percentile 8.4MB. A 15MB limit covers about 98% of them.
The limit is not about coverage but cost. Inlining images makes the output three to five times the size of the input, so a 40MB document becomes a response well over 100MB. If your file exceeds the limit, removing or externalising images before converting is usually enough.
Page extraction is section-based, not page-based
This one surprises people, and it is a property of the format rather than a shortcoming of any particular tool.
HWP documents are organised into sections — runs of content sharing page setup — not into stored pages. Page breaks are a result of laying out a section. So requesting "pages 1-2" operates on section boundaries, and many real documents, press releases especially, are a single section covering the whole file.
When that happens the extraction cannot narrow anything down. Our converter detects it and says so, rather than returning the whole document as though it had been trimmed. HWP vs HWPX has more on the section model.
Choosing an output format by what you need
- Reading and printing — PDF. Closest to the original, opens everywhere.
- Editing — Word (.docx). Structure survives; expect to fix some spacing.
- Translation or AI summarisation — Markdown. Table structure is preserved explicitly, which plain text loses. See getting HWP text into a translator or AI tool.
- Searching or archiving — plain text. Smallest output, no formatting to maintain.