Grafting technique was used to estimate a forecasting model for yield (kg/ha) of sorghum in Nigeria. The estimated model was compared with the traditional models to show that time series data does not always relate linearly to trend over the entire sample space. The results show that the grafted model was superior to others.