The output power of photovoltaic（PV） is uncertain. In order to mitigate the negative impacts of the uncertainty on power grid， a grey and neural network（grey-NN） hybrid model is proposed to forecast the short-term output power of PV. This hybrid forecasting method combines traditional direct and indirect forecasting methods. It considers the main factors that influences the PV output power and builds the hourly power grey models by choosing proper samples which have the similar weather condition with the prediction day. Then a neural network model is set up to obtain the final predicted power. The input data of the neural network model are output of grey models plus temperature. This paper uses practical output power data of PV to evaluate three different models of grey， neural network and grey-NN. The evaluation results show that the grey-NN model can predict the short-term output power of PV more precisely and has a potential value in practical applications.