Somewhere in an archive sits the map you actually need: the 1890 survey of the estate, the wartime sheet of the harbor, the planning drawing of the neighborhood before the highway. It is rich, precise, and useless to your GIS, because as a scan it is only pixels, with no idea where on Earth it belongs. Georeferencing is the ceremony that fixes that. In QGIS's Georeferencer you place ground control points, GCPs, on the scan, each tied to real coordinates taken from a modern map, and the tool computes a transformation that drapes the picture over the world. Done well, the old sheet overlays today's streets closely enough to trace vanished buildings and compare a century of change. Done carelessly, it produces a rubber sheet that agrees at your points and lies everywhere between them. This guide teaches the careful version. A sibling article, how to georeference an image in QGIS, covers the same ground from a first-timer's angle; this one concentrates on the judgment calls, point placement, transformation choice, and honest validation, that decide whether the result deserves trust.
Home / GIS and Open Map Data for Beginners
The QGIS Georeferencer: Pin a Scanned Map to the Real World

Photo: Ylanite Koppens / Pexels
Controls Train, Checkpoints Judge: The Georeferencer Checklist
Check off each step as you finish it. Your progress stays on this page.
0 of 8 complete
0% done
Ground Control Points: Spread Them, Choose Them Well
Open the Georeferencer from the Layer menu, load your scan, and the work becomes a treasure hunt repeated a dozen times: find a feature on the old map whose modern position you can identify exactly, click it, and enter its real coordinates, either by typing them or by clicking the corresponding spot on a modern canvas you have prepared with a basemap and reference layers. The quality of the whole georeference is decided by two properties of your point set. Identity: each point must be the same place in both worlds, a road intersection that has not moved, a church corner, a bridge abutment, a survey marker; features that migrate, riverbanks, coastlines, forest edges, are traitors. Prefer points that are sharp at the scan's resolution, a crossroads reads to the meter, a town's general center reads to the hundred meters, and your result can never be sharper than your vaguest point. Spread: points must surround and fill the area you care about, corners first, then edges, then interior, because a transformation interpolates politely inside the fence of your points and extrapolates wildly outside it. A cluster of twelve points in the town center leaves the map's margins to guesswork, and the margins are usually where the interesting change happened. Gather more points than the transformation's minimum, spread them like you are pitching a tent and want the canvas taut everywhere, and record a few extra as checkpoints, a concept the final section makes essential.
Choosing the Transformation: Linear, Polynomial, or Spline
The transformation is the mathematical rule that converts your points' agreement into a whole-image warp, and the Georeferencer offers a family in increasing flexibility, which is to say increasing power to fit and to deceive. Linear is the humblest: it moves, scales, and rotates the image without warping it at all, and it writes its result as a world file beside the untouched image rather than producing a new raster. For a modern, dimensionally honest scan, a digital map exported from another program, linear is often exactly right, and its refusal to bend is a virtue: it cannot invent agreement. Polynomial transformations, first through third order, bend the image progressively, correcting the classic sins of scanned paper: uneven shrinkage, a scanner's slight skew, a sheet photographed at a mild angle. Second order is the working default for historical paper maps. Thin Plate Spline is the strongest medicine: it warps the image so it passes through every control point exactly, locally stretching the sheet like rubber pinned at your points. That perfection is seductive and dangerous, because the spline guarantees agreement where you measured and promises nothing anywhere else; between points it can bulge and sag in ways no paper ever did. The discipline is to use the least flexible transformation that brings residuals to an acceptable level, escalating only when the residuals themselves argue the image truly is bent, and to remember that every added order spends your points' evidence on flexibility. Resampling choice rides along: nearest neighbor preserves the original pixel values and the map's crisp lines, while smoother methods prettify at the cost of invention, so for map scans nearest neighbor is the honest default.

Residuals and Checkpoints: Trust, but Verify
As you place points, the Georeferencer reports each point's residual, the distance between where the transformation puts it and where you said it should be, in the target units. Residuals are your instrument panel, and they deserve a skeptical reading. A single large residual among small ones usually indicts the point, not the model: you clicked the wrong corner, misread a coordinate, or chose a feature that has in fact moved since the old map was drawn. Fix or delete that point and watch the field improve. Uniformly mediocre residuals across the sheet suggest the model is too stiff for the paper's real deformation, the argument for stepping up an order. Beautifully tiny residuals under a spline prove almost nothing, because the spline was built to produce them. That is why the checkpoints you held back matter: points measured like control points but not fed to the transformation, whose residuals are then an independent exam the warp did not study for. Judge the georeference by its checkpoints, not its controls. A result whose checkpoints land within your tolerance, a few meters for a town sheet, tens of meters for a small-scale historical map, across the whole area of interest, has earned overlay work. One whose checkpoints wander has told you the truth about itself, which is the most valuable thing a tool can do. Set the target CRS deliberately before you run, save the points file with the project so the work is reproducible, and write down the transformation and the checkpoint results; a georeferenced raster without its provenance is an opinion with the evidence thrown away.
Preparing the Scan: The Work Before the First Point
The georeference is largely decided before the Georeferencer opens, by the state of the image you feed it. Scan flat and square: a sheet fed through a scanner at a slight angle hands the transformation a skew it must spend its flexibility correcting, flexibility you wanted for the paper's own sins. Scan at a resolution matched to the source's detail, enough that the finest line you must click, a boundary, a building corner, is several pixels wide; a blurry scan caps the accuracy of every point placed on it. Trim the scan to the map itself where you can, since margins, legends, and neighboring sheets offer no control points and tempt misclicks, but never trim away the sheet's own marginal information before recording it, because the old map's imprint, its survey date, its original projection and datum notes, is provenance your result will be judged by. That last point deserves emphasis: many historical sheets were drawn on datums and projections that differ from today's, and while a polynomial warp absorbs the difference silently, knowing the source's stated system, where it survives, tells you what size of disagreement to expect and keeps you from "correcting" a map that was honestly drawn on its own terms. Prepare the modern side with equal care: a canvas holding a current basemap plus any authoritative reference layers, boundaries, surveyed points, modern mapping of surviving features, so that every coordinate you hand the Georeferencer is itself well-sourced. Garbage coordinates in, warped garbage out, with excellent residuals.
After the Warp: Using a Georeferenced Map Honestly
The output raster is the beginning of the work, not its end. Loaded into your project, the old sheet should be drawn at partial transparency over the modern map for its first inspection: coastlines and road nets should embrace, and the places they refuse to are a map of your result's local honesty, more informative than any single residual. Digitize from it with its error in mind: a building traced from a georeference whose checkpoints wander by fifteen meters is a fifteen-meter claim, and features that must be exact, boundaries with legal weight, deserve modern sources, with the old map as evidence of history rather than of position. Keep the family together in your files: the scan, the points file, the warped output, and a note of the transformation, the target CRS, and the checkpoint results, because a georeferenced raster divorced from its provenance cannot be re-examined, only believed. And expect the work to be iterative across a project: the first sheet teaches you the archive's habits, its paper, its surveyors, its datums, and the fifth sheet georeferences in half the time at twice the trustworthiness. That accumulation, a whole archive brought back into geographic conversation with the present, is the real prize the Georeferencer offers, and it is worth the discipline this article has insisted on at every step.
Related Guides
Readers working through this topic usually need its closest companions too. These GIS and open map data guides cover the neighboring ground:
- QGIS Print Layout: Compose a Map Someone Can Actually Use: The map on your QGIS canvas is a working view; the map in a print layout is a finished argument, with a fixed scale, a legend that explains it,...
- Add a Point by Coordinates in QGIS: Exact Placement With the Digitizing Panel: Clicking a point onto a map is an estimate wearing a costume.
- How to Add a Shapefile in QGIS: The Whole File Family, the CRS Prompt, and the Zip Trick: Adding a shapefile sounds like opening a file, and it is, once you understand that a shapefile is not a file.
Frequently Asked Questions
What does the QGIS Georeferencer do?
It aligns a raster image, such as a scanned historical map, to real-world coordinates. You supply ground control points tying features on the image to known coordinates, and the tool computes a transformation that positions, and if needed warps, the image so it overlays modern data correctly.
How many ground control points do I need?
More than the transformation's minimum, spread across the whole image: corners, edges, and interior. A linear transformation needs at least two points mathematically but deserves several; polynomial work wants a dozen or more. Spread matters as much as count, because accuracy collapses outside the fence of your points.
Which transformation should I choose?
The least flexible one that fits. Linear for dimensionally honest digital images, writing a world file without warping. Polynomial, usually second order, for scanned paper with real shrinkage and skew. Thin Plate Spline only when local deformation is genuine, since it forces exact agreement at control points and can distort everywhere else.
What is a residual, and what is a checkpoint?
A residual is the error at a control point after transformation, the gap between the point's mapped and computed positions. A checkpoint is a point you measure but withhold from the transformation, so its residual is an independent test. Judge a georeference by its checkpoints, not by the controls the transformation was fitted to.
Why does my georeferenced map fit at the points but not between them?
The transformation is interpolating beyond its evidence, usually because the points are clustered, too flexible a warp is bending the sheet unnaturally, or the source map itself was drawn with local inaccuracies. Add well-spread points, consider a stiffer transformation, and check with withheld checkpoints where the error truly lives.
Photo credits: Ylanite Koppens (Pexels); Ylanite Koppens (Pexels). Photos are served from this site; follow the credit to the library page for the original.