In this paper we propose a new gene network reconstruction (or identification) scheme which takes advantage of the sparseness of a gene network using a decomposition of the given linear dynamical system describing the network, into two positive linear systems. First, we will describe how gene networks can be modelled as linear systems and an “ideal” situation is considered in order to state an identification problem for gene regulatory networks. Finally, some preliminary results on the algorithm performances obtained using artificially generated data will be presented.