There are many sources of inaccuracy, or error, in a survey. Sampling error consists of bias in sample selection procedures, plus random sampling error. Non-response bias can be measured and analysed after interviewing is completed. There is scope for less visible and less measurable error in the interviewing process itself, and in the subsequent coding and classification of replies. Interviewer bias affects some interviews, and interviewers do occasionally make mistakes, such as overlooking a whole section of a questionnaire. Coding errors arise when data are prepared for analysis. They consist of simple mispunches, from striking the wrong key to code a reply, and misclassification, as when a job description is not read or understood correctly, and is allocated to the wrong occupation code. Edit and consistency checks after data preparation will identify some but not all coding errors. Surveys require rigorous attention to detail at every stage in the process to reduce error to the minimum. Even small inaccuracies at each stage can mount up to an appreciable amount of total error in the end.