The behavior of postwar real U.S. GNP, the inputs to an aggregate production function, and several formulations of the associated Solow residuals for the presence of nonlinearities in their generating mechanisms are examined. Three different statistical tests for nonlinearity are implemented: the McLeod-Li test, the BDS test, and the Hinich bicovariance test. We find substantial evidence for nonlinearity in the generating mechanism of real GNP growth but no evidence for nonlinearity in the Solow residuals. We further find that the generating mechanism of the labor input series is nonlinear, whereas that of the capital services input appears to be linear. We therefore conclude that the observed nonlinearity in real output arises from nonlinearities in the labor markets, not from nonlinearities in the technical shocks driving the system. Finally, we investigate the source of the nonlinearities in the labor markets by examining simulated data from a model of the Dutch economy with asymmetric adjustment costs.