Investments in lifelong learning may create unsatisfactory results, and this could potentially contribute to the reproduction of inequalities. We argue that the process is related to the accumulation of opportunities and barriers for participation in training, which can lock individuals in disadvantageous path-dependent trajectories. We take a longitudinal approach to analyse whether participation in training in older age is path-dependent, and whether this path dependency is related to institutional contexts. Using data from the Survey of Health, Ageing, and Retirement in Europe (SHARE), we trace individual training trajectories in the population aged 50+ in twelve European countries between 2010 and 2015 (27 370 respondents). Hierarchical Bayesian logit models serve to assess the probability of training during the sixth wave, with a lagged dependent variable as a predictor. Results suggest that training participation is path-dependent and participation in training is limited for people who have not trained previously. It is also related to macrostructural context: path dependency is lower in countries with stronger knowledge economies, stronger emphasis on education, and a proactive ageing climate. Recognising path dependency can help to improve access to training and design policies that address problems of cohesion, active ageing and adult learning.