The dissertation investigates the human capital accumulation through on-job learning. It has three empirical studies. The first two essays investigate skill accumulation through performing job tasks. The third one analyzes the labor market effect of tertiary education for full time workers. Using the data from the Programme for the International Assessment of Adult Competencies (PIAAC) for USA, we firstly investigate whether cognitive skills can be improved through on-the-job learning, especially via tasks at work. With rich information on job tasks performed at individual level, we construct different job complexity measures: a general job complexity measure, and two specific complexity measures of interactive and analytical tasks. The results show that workers can accumulate cognitive skills through solving complex problems. Additionally, analytical tasks play an important role on cognitive skills, while interactive tasks at work do not show a significant effect. Furthermore, we investigate whether tasks performed at work contribute to the improvement of a worker’s problem-solving skills. Based on two datasets for Germany, the Programme for the International Assessment of Adult Competencies (PIAAC) and “LLLight’in’Europe” project (LLL), we analyze on problem-solving skills at different levels, general problem-solving skills and complex problem-solving skills. The results of two problem-solving skill measures show workers benefit from doing a complex job, and task complexity improves complex problem-solving skills with a much smaller magnitude. In addition, analytical tasks at work play a more important role than interactive tasks. The third essay investigates the difference in effects of tertiary education between full-time workers and full-time students, based on data from the Chinese Household Income Project (1995, 2002, 2007, and 2013). We find that the schooling returns to a college and a graduate degree earned by full-time workers are significantly lower than the returns on corresponding degrees earned via full-time studies, however, there is a much smaller or no significant gap for junior college degrees for those two groups. The results are quite robust with different model specifications and estimation methods. Our further investigation shows that school quality or aging cannot explain the gap fully.