This paper addresses the problem of data mining from temporal data based on calendar (date and time) attributes. The proposed methods uses a probabilistic domain generalization graph, i.e., a graph defining a partial order that represents a set of generalization relations for an attribute, with an associated probability distribution for the values in the domain represented by each of its nodes. We specify the components of a domain generalization graph suited to calendar attributes and define granularity, subset, lookup, and algorithmic methods for specifying generalizations between calendar domains. We provide a means of specifying distributions. We show how the calendar DGG can be applied to a data mining problem to produce a list of summaries ranked according to an interest measure given assumed probability distributions.