Abstract. This paper analyzes the demographic characteristics of unemployment in Spain using individual household data. Binary‐logit models are estimated for three microdata sets at different periods of time, in order to detect possible compositional changes in unemployment over time. The results indicate that characteristics such as age, education level and marital status are relevant to explain differences in unemployment. The fact that the male and female models turn out to be very similar may be due to hidden female unemployment. The estimated models are used to predict unemployment probabilities as a function of individual characteristics.