The return generating process of the arbitrage pricing theory is examined using factor analysis. Over the period 1953-1971 not only do formal statistical tests fail to confirm the notion of industry factors, but the statistics suggest that the number of factors may be large. In addition, factor analysis seems to be sensitive to the choice of assets included in the covariance matrix. For example, analyzing bonds and stocks together reveals additional factors, common to both groups, which are not found when only one set of securities is examined. Finally, factor analysis assumes that the covariance (or the correlation) matrix is constant over time. Such an assumption is not necessarily consistent with the data.