A new prediction technique is proposed for chaotic time series. The usefulness of the technique is that it removessome false neighbouring points which are not suitable for the local estimation of the dynamics systems. We use afeedforward neural network to approximate the local dominant Lyapunov exponent, and choose the neighbouringpoints by the exponent. The model is tested for the convection amplitude of the Lorenz model, and the resultsindicate that this prediction technique can improve the prediction of chaotic time series.