When dealing with continuous data, a typical assumption is that the continuous/real-valued features associated with each class are distributed according to a normal (or Gaussian) distribution. Statistics computed by segmenting the data by the class, and then computing the mean and variance of the continuous attributes in each class.