Inference about regression models, like standard errors and confidence sets, are dependent on assumptions about the true definition of the model. Accurate and reliable calibration results based on ordinary least squares (OLS) regression will only be obtained, if the assumptions of OLS are actually fulfilled. This parr of the report starts off with reviewing statistical techniques To validate the calibration data, i.e., to check if the data are compatible with the assumptions on which OLS rests. In the following regression techniques are reviewed that are more efficient than OLS, in case one or more of the classical assumptions are violated. Included are generalized least squares with variance function estimation, transformations of the response and/or predictor variables, and robust regression. Finally, a brief review about the application of ＇The Bootstrap＇ in analytical chemistry is provided, which is a conceptually easy, though computationally intensive, alternative for estimating standard errors, confidence intervals, and bias of a statistics by resampling the data in a suitable way. An extensive bibliography is provided.