Home / GIS and Open Map Data for Beginners

Geocoding in QGIS: Turn a Table of Addresses Into Points With MMQGIS

Aerial view of a busy urban intersection in autumn

Photo: Chandra Phuyal / Pexels

Geocoding is the alchemy that turns the most common dataset in the world, a table of addresses, into the most useful one, a layer of points. A housing charity has shelters as a spreadsheet. A delivery firm has customers as a spreadsheet. A historian has a century of addresses in a ledger. None of them is a map until each row acquires coordinates, and acquiring them at scale, hundreds or thousands of rows at a time, is a job for a geocoding service driven from inside QGIS. The desktop workflow runs through a long-standing plugin, MMQGIS, which sends your address fields to a geocoding service and builds the point layer from what comes back. This guide covers the whole arc: preparing the table so the machine can read it, running the geocode, and then the part that separates a layer you can publish from a layer you merely have, auditing what came back, including the addresses the service could not find. The service side of this world has rules, and our Nominatim guide explains the OpenStreetMap geocoder and its usage policy in depth; this article is about the QGIS side of the conversation, the desktop workflow that feeds it.

Interactive tool

Addresses In, Points Out: The QGIS Geocoding Checklist

Check off each step as you finish it. Your progress stays on this page.

0 of 7 complete

0% done

Before You Geocode: The Table Is Half the Job

Geocoders read structure, not intentions, and the quality of the output is decided before the plugin ever runs. Your table needs the address in separate columns: street address in one, city in another, state or province in a third, country in a fourth, because the QGIS workflow maps your columns to the service's fields, and a single merged address column forces the service to guess where the street ends and the city begins. Clean the data as if a stranger had to deliver a parcel from each row, because in effect one does. Spell city and street names the way the map spells them, not the way the office abbreviates them. Remove the rows that are not addresses at all, the headquarters that is really a post office box, the client listed as various locations, the duplicate that appears twice with different punctuation, because each will either fail or, worse, succeed wrongly. Keep an ID column that ties each row back to your source data; the geocoded layer will gain coordinates, and the ID is how you prove, later, which point was which row. Save the cleaned table as a CSV, the format the workflow consumes. This preparation feels like delay and is actually the geocode itself, performed by the only participant in the process who knows what the addresses mean.

Install MMQGIS and Run the Geocode

MMQGIS installs like every QGIS plugin: Plugins, Manage and Install Plugins, search, install, as our plugins guide describes. Its geocoding lives at MMQGIS, Geocode, Geocode CSV with Google / OpenStreetMap, a dialog that asks for three things: your CSV file, the service to query, and the mapping between your columns and the address fields the service expects. Walk the mapping slowly. Your street column goes to the address field, your city to city, your state or province to state, your country to country; the dialog exists because no two spreadsheets name these things alike, and a column mapped to the wrong field produces confident nonsense at industrial speed. Choose the output layer's name and location, and note the second output the dialog offers, the Not Found file, and make sure it is being written somewhere you will look. Then run it, and be patient in proportion to your table: services answer one request at a time at a polite pace, and a thousand rows is a coffee, not a click. When the run completes, the found addresses appear as a point layer over your project, and the work of the next two sections begins, because a geocode that has not been audited is a rumor with coordinates.

High-angle view of a busy urban intersection with skyscrapers and traffic
High-angle view of a busy urban intersection with skyscrapers and traffic

The Not Found File: Where the Honesty Lives

Every geocoding run produces two outputs, and the smaller one matters more. The Not Found file lists the addresses the service could not place, and it is the difference between a workflow and a hope. Read it as a diagnostic, because its failures cluster into kinds, and each kind has a fix. Misspellings and local abbreviations the map does not use: correct them to the map's spelling and rerun just those rows. New developments and renamed streets, where your table is newer or older than the map: these may need a manual placement, by the methods of our add a point by coordinates guide once you establish where they are. Addresses that were never addresses, the post office boxes and various locations you meant to clean out: delete them with a clear conscience. And a residue that is genuinely ambiguous, two towns sharing a name, a street that exists in both: resolve them by adding the disambiguating detail, a district, a postcode, a neighboring landmark town, to the row. Rerun the corrected failures as their own small batch rather than re-geocoding the whole table; the services' usage policies reward not asking twice for what you already have, and so does your afternoon. A finished project can state its match rate plainly, this many placed, this many placed by hand, this many unresolvable, and that statement is what makes the layer professional.

Check the Points Before You Trust Them

A placed point is not necessarily a correct point, and geocoding has a characteristic failure that every user meets: the fallback. When a service cannot find the exact address, it often returns the best thing it has, which may be the street's midpoint, the town's center, or a same-named street in the wrong district, and it returns it with the same confidence as a rooftop match. The audit that catches this is visual and fast. Lay the points over a basemap and look at the pattern before you look at any single point: points should follow the geography of your table, clustering in the towns your addresses name, strung along the streets they claim. A point sitting in a harbor, a point alone in a field while its hundred siblings cluster in a city, a neat stack of points at one identical spot, which is the town center fallback wearing a crowd as camouflage: these announce themselves to the eye in seconds. Then spot-check a sample against local knowledge, the addresses you personally know, the office, the famous landmark, your own street. Geocoding quality also varies with the address's own quality and the map's coverage of its country, which is why the audit is not paranoia but calibration: you are learning how much this run, on this data, in this place, deserves to be trusted, and the answer decides what the layer may be used for.

The Other Doors: One-Off Lookups and Local Geocoding

MMQGIS batch geocoding is the heavy tool, and QGIS offers lighter ones for lighter jobs. For a handful of addresses, the OSM Place Search panel and single-address geocoding plugins query the same services interactively, one lookup at a time, with the result dropped on the map for immediate inspection; for five addresses, that is the right scale of tool. For work that must not leave the building, or addresses in a jurisdiction with an authoritative street database, street-layer geocoding matches your table against a local street network layer instead of a web service, interpolating positions along named street segments; it is more setup, and it is the method of record for much official work. Reverse geocoding runs the whole machine backwards, coordinates in, nearest address out, and MMQGIS and the service panels both offer it, for the fieldwork dataset that arrives as points and needs its addresses named. Choose by the shape of the job: batch service geocoding for the spreadsheet, interactive lookup for the handful, local geocoding for the authoritative or the private, reverse for the points that came first. All four share the audit discipline of the previous sections, because all four can be wrong in the same quiet ways.

The Rules of the Road: Services Are Shared

A geocoding service is a commons, and the workflow's manners are part of its method. The public Nominatim service, whose policy our Nominatim guide sets out, asks for at most one request per second, forbids autocomplete-style hammering, requires you to identify your application, and expects you to cache results rather than ask again for addresses you already placed. Commercial services carry their own quotas and terms, which change over time and differ by plan, so read the current terms of the service you choose rather than trusting any number printed in a tutorial, including the ones in older tutorials for this very workflow, whose figures date quickly. The practical habits follow from respect and self-interest alike: geocode once and keep the result, run the failures as small correction batches instead of re-running the table, and do the heavy runs at a scale and pace the service invites. A layer built politely is also a layer built reproducibly, because a workflow that depends on abusing a free service is a workflow that stops working the day the service protects itself, which they all, eventually, do.

Match Quality: Rooftop, Street, or Town

Not all placed points are equal, and professional geocoding keeps the distinction the services themselves draw. A rooftop match places the point on the building the address belongs to, the gold standard, fit for navigation and delivery. A street match places it along the right street, interpolated between known numbers or set at the segment, fit for mapping and most analysis. A locality match, the town or district center fallback described earlier, places it in the right settlement and the wrong everything else, fit for counting addresses per town and for nothing finer. The services know which grade each result earned, and the workflows differ in how loudly they tell you; the audit of the earlier section is how you find out when the software stays quiet. The discipline is to let the grade govern the use. A layer of rooftop and street matches can guide a driver. A layer salted with locality matches can still serve a regional study, provided the salt is measured and disclosed. What no layer may do is wear a uniform confidence it has not earned, which is why the match rate this article keeps mentioning has a companion: the match quality distribution, stated alongside it, so the map's reader inherits your knowledge of its limits rather than your silence about them.

Related Guides

The closest companions to this guide are the GIS and open map data guides below. Each one picks up a thread this page starts:

Frequently Asked Questions

Which geocoding service should I choose in MMQGIS?

The dialog offers the Google and OpenStreetMap services. The OpenStreetMap route, through Nominatim, is free and open, with a usage policy that suits moderate batches and demands caching and a one request per second pace. The Google route follows Google's current terms and quotas. Choose by your volume, your budget, and the terms you can honestly keep.

Why did so many of my addresses land in the same spot?

That is the fallback signature. When the service cannot find an exact address, it returns a coarser match, often the town or street center, so many failures pile onto one point. The Not Found file and a basemap audit reveal them; fix the addresses that can be fixed, with better spelling or more detail, and rerun only those.

What do I do with addresses the geocoder cannot find?

Work the Not Found list by kind: correct spellings to the map's forms, add disambiguating detail like a district or postcode, delete rows that were never addresses, and place the genuinely new or renamed locations by hand once you establish their coordinates. Report the final match rate honestly with the layer.

Can I geocode without sending my addresses to a web service?

Yes. Street-layer geocoding matches your table against a local street network dataset inside QGIS, so nothing leaves your machine. It takes more preparation, and it is the standard approach for private data and for jurisdictions with an authoritative street database.

Is it acceptable to geocode the same table again next month?

Re-running the whole table wastes the service and your time, and good practice is to cache: keep the geocoded layer, geocode only the new and changed rows, and merge them in. The services' policies expect exactly this, and your match-rate bookkeeping stays clean when each row is placed once.

Photo credits: Chandra Phuyal (Pexels); Willian Justen de Vasconcellos (Pexels). Photos are served from this site; follow the credit to the library page for the original.