Skip to content

Stockertim Dictionary · Technology

Database normalisation

Also called database normalization, data normalisation, normal forms, third normal form, 3NF

In short

The design practice of splitting data into linked tables so that each fact is recorded in exactly one place, which stops copies of the same fact disagreeing.

  • Intermediate
  • Process
  • Checked

In detail

Normalisation is a set of design steps for relational databases. Its aim is that every fact appears once. If a supplier's phone number sits on every purchase row, a change has to be made on hundreds of rows, and any row missed now disagrees with the rest. Moving suppliers into their own table, and keeping only a supplier code on each purchase, removes the repetition.

The steps are described as normal forms. First normal form asks that each cell hold a single value, so a column listing three phone numbers separated by commas is split out. Second normal form removes details that depend on only part of a combined key. Third normal form removes details that depend on another non-key column, such as a town name that follows from a postcode stored in the same row. Most business systems are designed to about third normal form.

The problems normalisation prevents are known as update, insertion and deletion anomalies. An update anomaly is the half-changed phone number. An insertion anomaly is being unable to record a new supplier until it has made a sale. A deletion anomaly is losing the only copy of a supplier's details when its last purchase is removed.

Normalised tables need joins to answer everyday questions, which costs some speed and adds steps. Reporting databases and analysis files are therefore often denormalised on purpose: a flat table repeats descriptive values so that summaries are quick. That is a sound choice for reading, provided the master copy that people edit stays normalised.

Worked example

A furniture maker in Leeds keeps 3,000 purchase rows, each repeating the supplier's name, phone number and town for one of 40 suppliers.

  1. Repeated supplier fields before: 3,000 rows × 3 fields = 9,000 cells
  2. After moving the details to a suppliers table: 40 rows × 3 fields = 120 cells, plus one supplier code on each purchase row

A supplier's new phone number is changed in one cell, and no purchase row can show an out-of-date number.

In practice

Designers normalise when they build a system that people edit every day. Analysts meet the result as several linked tables and learn to join them back together, and they often build denormalised extracts for dashboards. Spreadsheet users apply the same idea when they keep a separate lookup sheet of codes instead of retyping descriptions.

Watch out

  • Normalisation is about where facts are stored, not about scaling numbers; the statistical sense of normalising values is a different idea.
  • A flat reporting extract is fine to read but should never become the place where corrections are made.

Learn it properly

Official sources

Facts last checked on 9 November 2026.

Education, not advice: the Stockertim Dictionary explains what terms mean. It never recommends buying or selling anything.

English