The traditional methods of statistical classification of text spam are analyzed. For a naive Bayesian classifier,
a number of modifications are proposed. Based on these modifications, an alternative algorithm is described
and its time and space complexity are characterized.