Occupations are bundles of inseparable skill requirements and tasks. We propose a novel approach for studying the relationship between wages and bundles of occupational skills and tasks. We predict occupational wages using a regression tree approach which also provides an empirically powerful aggregation scheme where detailed occupations with similar wages and job requirements are combined into 15 large occupation groups. Our empirical analysis is carried out on a dataset obtained by combining O*Net information on job attributes with the occupational wage and employment information from Occupational Employment Statistics. Not having a priori information on which O*Net variables belong in a wage equation, the first step in our analysis is to perform factor analysis on a number of O*NET categories that represent basic job skill requirements and job attributes. The second step is to use a regression tree to group the detailed SOC occupations into broader aggregates. These occupational aggregates are then used to non-parametrically analyze the well-known hollowing out phenomenon and the increase in log wage variance from 2005 to 2017.