Summary Survey questionnaires often contain open‐ended questions, for which interviewers are required to record respondents' replies verbatim. Errors can arise in coding these replies in preparation for statistical analysis. The paper reports the results of an experiment examining the levels of reliability attained by six professional coders in making judgemental codings of a sample of responses to six survey questions. A sizeable degree of unreliability was found, especially with the use of general and “catch‐all” codes. Intra‐coder correlation coefficients, measuring the correlated component of coder variance, were generally small; it is noted, however, that even small values of these coefficients can lead to a substantial loss in precision for survey results.