The absence of repeated direct measures of job skill requirements, such as a fully updated edition of the Dictionary of Occupational Titles (DOT), creates problems for anyone interested in understanding whether, how much, and how fast the demand for cognitive skills in the workplace has grown over time. Many studies have merged cross-sectional skill scores for detailed occupations from the DOT or O*NET with time series of occupational employment shares, but this fails to capture any within-occupation changes in job requirements. This paper considers the utility of using workers’ personal education as a time-varying measure of occupational skill requirements. Trends in educational composition within detailed occupations for 1990-2001 are examined using very large sample data. Shift-share analyses decompose the total change in workforce education into components attributable to changes in occupational employment shares and in education levels within occupations. Because occupations are often characterized in terms of modal education, the extent of educational heterogeneity within occupations is examined. Limitations of workers’ personal education as a measure of job complexity, the need for repeated direct measures of job requirements, and implications for BLS data programs are discussed.