[Classification of multidimensional observations using linear discriminant functions].

In the present paper the role of linear discriminant analysis is delimitated all along with a range of problems which could be solved using this method in appropriate conditions. The analysis is given of possibilities on how to involve the decrease in number of variables in a model, and how to predict probability of incorrect classification. As an example of application, the prediction of dog survival is calculated in radiobiological experiment on the basis of biochemical and hematological values which were obtained by the 2nd day after the irradiation as well as an error probability was determined.