Based on theoretical models of job mobility this paper provides an empirical analysis of job durations in West Germany using information from two cohorts of new entrants to the labor force. We adopt an accelerated failure time model allowing for unobserved heterogeneity. Thereby we combine the generalized estimating equations approach with the replacement of censored times by imputed values. The most important results are that employees stay longer in larger establishments and in production industries. Older individuals, those with completed apprenticeship training and those coming directly from apprenticeship training have longer job durations. In contrast, the larger the number of unemployment and employment spells, the shorter is the job attachment expected.