A deterministic approach is proposed for proving the convergence of stochastic algorithms of the most general form under necessary conditions on the input noise and reasonable conditions on the (nonnecessarily continuous) mean field, Emphasis is placed on the case where more than one stationary point exists, We also use this approach to prove the convergence of a stochastic algorithm with Markovian dynamics.