This article examines the reciprocal relationships between work variables and health outcomes and if these relationships differ by social class (measured by occupational grouping). We used longitudinal data from the 1994/95--2002/03 Canadian National Population Health Survey (NPHS). Karasek's work stress variables were measured in the 1994/95 (cycle 1, time 1), 2000/01 (cycle 4, time 2) and 2002/03 (cycle 5, time 3) surveys. Analyses were limited to 2556 respondents aged 18-56 at time 1 and who remained in the same social class (as defined by occupational position) for all the three time points. Work variables used were job strain ratio, work social support and job insecurity. Health outcomes used were distress, depression and self-rated health. Multi-group path analyses were used to investigate the reciprocal relationships between work and health variables and if these relationships differed by social class. Analyses controlled for age, gender, marital status and work status. We find there is a differential burden of work psychosocial factors and health outcomes by social class. The cross-lagged relationships between work and health depended on the outcome, social class and time lag. More significant paths from work to health were observed than reverse paths from health to work. More significant relationships between work and health were observed for the shorter time lag (2 years) compared to longer time lags (6 years). Low work social support and job insecurity were more detrimental to health for respondents in lower social class positions. Findings from this study highlight the importance of time lag, and to some extent social class, in the reciprocal relationships between work and health.