A longitudinal modelling approach is used to examine the effect of job satisfaction on survey non-response in an applied employee attitude survey. Members of a large electronics company were invited to participate in the survey in each of four consecutive years. Logistic autoregressive modelling revealed a significant effect of prior non-response on subsequent non-response. The odds of responding to the survey in any given year were more than twice as high for individuals who had responded to the same survey 1 year before, as compared with those who did not. Prior job satisfaction was also significantly related to subsequent survey non-response. Participants who were more satisfied on a given occasion were more likely to respond 1 year later than those who were less satisfied. At the facet level, satisfaction with supervisors contributed incrementally to the prediction of survey non-response in subsequent years. Implications for the theoretical and methodological development of non-response research and for the practical implementation of employee attitude surveys are discussed.