Methods of forecasting on time series data by auto-regressive model, neural networks model and chaotic theorem are proposed. The methods are said to be complicated for forecasting calculations. In this paper, a simple forecasting algorithm by the nearest neighbor method is discussed. It is applied to forecasting on a monthly average temperature in Sendai city to confirm performance. The actual results indicate that the nearest neighbor method has similar performance to the auto-regressive model in forecasting on time-series data.