A dynamic latent structure model of the work–retirement transition process was identified, focusing on transitions of work and retirement status for men and women aged 51–74 years. Using the Health and Retirement Study data (1998–2004), latent transition analysis was used to identify a best fitting model capturing work–retirement statuses in four samples defined by age and sex. The prevalence of each status was described and the dynamic transition probabilities within the latent structure were examined. Using multinomial logistic regression, socio-demographic, health, family and occupational factors were assessed to determine how each was related to the likelihood of occupying a specific latent status at baseline. Results showed that study respondents were classified into distinct groups: full retiree, partial retiree or part-time worker, full-time worker, work-disabled or home-maker. The prevalence of full retiree status increased, while the prevalence for full-time worker status decreased over time for both men and women. Membership rates in the work-disabled and partial retiree status were generally consistent, with decreased probabilities of the work-disabled status in the older age groups and increased probabilities of partial retirees among younger men. Our findings indicated that many older Americans experience multiple transitions on the pathway to retirement. Future research on late-life labour-force transitions should evaluate the impact of the recent Great Recession and examine the role of larger socio-economic contexts.