In nonlinear regression analysis, measures of curvature were proposed by D. Bates and D. Watts as a means of assessing the reliability of certain linear approximations. Curvature has two components: parameter-effects curvature, which can be reduced by suitable transformations of the parameters, and intrinsic curvature, which represents the essential aspect of nonlinearity. Estimation of these measures requires the second derivatives of the expectation functions in the model.