The paper presents farther results on the problem of identification of a minimal order dynamic model for the behaviour of the active load demand at a number of nodes of an interconnected power system. An earlier paper contains the results of application of the vector time series techniques. In the present paper a stochastic state variable model is identified using the recently proposed algorithm of Tse and Weinert (1976). Real data pertaining to a segment of the northern India power grid have been processed using this algorithm, and short-term load forecasts obtained through the state variable model are compared with those based on the time series model.