This paper develops an error components model that is used to examine the impact of job changes on the dynamics and variance of individual log earnings. I use data on work histories drawn from the Panel Study of Income Dynamics (PSID) that makes it possible to distinguish between voluntary and involuntary job-to-job changes. The potential endogeneity of job mobility in relation to earnings is circumvented by means of an instrument variable estimation method that also allows to control for unobserved individual-job specific heterogeneity. Once controlled for individual and job-specific effects, the persistence within jobs is almost zero, whereas across jobs is significant but small.