A reason of the lack of agreement between measured pollutants concentration in the air of urban areas and vehicle pollutant emissions evaluated by available emission models is the fact that catalyst performance variability is not considered. In this paper, an experimental study on the effect of performance variability of catalyst on emissions is presented. Average emissions have been measured using driving cycles representative of different levels of urban traffic, determined by statistical methods on the basis of data detected on-road by an instrumented car. For each driving cycle, representative of a certain traffic level, different thermal starting conditions of catalyst have been tested. These conditions have been determined by the characterization of catalyst performance at steady state and are representative of real catalyst conditions experienced on the road. In particular, the combined effects of real driving cycles (rapid acceleration and deceleration, short trips in which the engine never fully warm up, most time in congested traffic with engine at idling) and of catalyst thermal behaviour on emissions have been investigated. The final aim of the study is to develop a probabilistic approach for emission modelling in which emissions produced for a specific driving cycle in a certain urban zone are predicted on the basis of possible routes to reach the site where the cycle is to be performed.