The new method treats the problem of parameter estimation and noise reduction for chaotic time series as a nonlinear minimization process and solves it using steepest descent algorithm.