Net-driven decomposition techniques are considered in this paper in order to reduce the state explosion problem for the computation of performance indices of stochastic Petri nets.Basically, the idea is to represent (or partially represent) in a decomposed manner the reachability graph of the model so it can be used for exact and/or approximated performance analysis. In that way, the complete storing of the graph is avoided and, for the case of approximate analysis, the solution of the isomorphous continuous time Markov chain is substituted by the solution of smaller components. The techniques are applied to a couple of non-trivial models.