A new stochastic approximation algoritm is considered for the problem of optimal dissection of a distribution, which is only observable by a random variable and asymptotic properties are indicated. Under some conditions, the process converges to a point of minimum of the disconcentration funtion G(Q.