In the face of rapid technological change, a growing body of research has argued that the key determinant of employment growth is the nature of the tasks performed on the job and the extent to which they are complementary to IT devices. However, to make inferences about the task or skill content of jobs, previous studies often rely on standard occupational classifications, thereby implicitly assuming that occupations are made up of well-defined and homogeneous ``bundles of tasks''. In this article, we evaluate this assumption using a unique dataset of 60 million online job postings in the UK. Based on the skill requirements of job postings, we map the skill structure of the labor market using topic modeling and analyze its relationship to existing occupational classifications. Our results show that the labor market is segmented into coherent bundles of skills, but that these largely cross occupational boundaries. These results challenge the usefulness of occupations as proxies for skills and offer new perspectives for the analysis of labor market stratification.