The aims of this project were to investigate the signal processing and the experimental design requirements to objectively select the optimum combination of samples and variables that facilitate the development of a robust calibration when only small data sets are available and to identify mathematical tools that enable the user to utilise only a minimum number of samples. The definition of a small data set can be expressed as : i) the selection of the smallest number possible of standards or samples on which to build a representative and robust calibration and ii) the selection of the smallest number of measurement variables that minimise the prediction error. An additional aim was to develop methodology to allow for the selection of appropriate samples for modelling from a large or historical database of existing samples, and then the subsequent updating of that model. Secondary issues investigated were appropriate criteria for the selection of the number of latent variables to be used in a model, and the use of alternative modelling methods such as Ridge Regression, Evolving Window Factor Analysis, Curve Resolution techniques and Digital Filters.