An ensemble of mixture density neural networks is used for short-term wind speed and power forecasting.m Predicted wind speeds obtained from a numerical weather prediction model are used as the input data for the mixture density network, whose outputs are the mixture density parameters (used to represent the probability density function of the uncertain output or target variable). All mixture density neural networks in an ensemble are assumed to have a three-layer architecture, with each architecture having different numbers of nodes in the hidden layer. Because a mixture of Gaussian distributions is used to approximate the conditional distribution of the target random variable (either wind speed or wind turbine power), the uncertainties arising from both the model structure and model output can be completely quantified. In consequence, rigorous confidence intervals reflecting these sources of uncertainty in the prediction can be obtained and used to assess the performance for the wind speed and wind turbine power forecasting. An application of the proposed approach to a data set of the measured wind speed and power from an operational wind turbine in a wind farm in Taiwan is used to test the methodology. The results of this application demonstrate that the proposed methodology works well for the multi-step ahead wind speed and power forecasting.