Multi-step-ahead frocesting is the task of predicting a sequenceof some future values, given its past observations. Several methods have been used for predicting the nextfuture values, namely the iterative, direct and parrarel approach. There are also several algorithms that have been proposed to do such forecast. Meanwhile, cross- validation is a well known method to perform classification and regression tasks. This paper aims to implement the cross-validation method to forecast the result with other approaches. Our hypotheis is that cross-validation would provide better generalization since it does not use the entire data set in the training phase, hence, it already learns the unseen data during the training. The algorithm used to make prediction is Neural Network as it is widely used in literature for time series tasks. The dataset used in the experiment is the time series data designated for NN3 Competition. The experimental result shows that the cross-validation yields better performance, which is around 20-30 percent better, than iterative and direct approaches. In addition, the esperimental result indicates that the optimal number of input sequence is abbout 25 percent of the number of training samples. Keywords-time series; forecasting; predicition; multi-step-ahead; cross validation; neural network