In this paper we present a rule-based system of automatic analysis of temporal relations within a discourse for Czech. The core algorithm of the system is the so-called "recursive temporal property" (RTP) of Czech that predicts relative ordering relations between finite verbs in a sentence based on the information provided only by the grammar. We correct and extend the RTP by various linguistically-oriented rules. Some of these rules represent a contribution to the existing grammar theory of temporal relations in Czech. The resulting system yields F-measure of 57.77-65.29.