According to the complexity, diversity and uncertainty of the influence factors on groundwater depth change, and the complicated nonlinear relation between these factors, the adaptive neuro-fuzzy interference system(ANFIS) was put forward for predicting the groundwater depth. By training the ANFIS according to the irrigation materials of Jinghuiqu irrigation district in Shaanxi Province during 1993-2010, the prediction model for groundwater depth was built based on ANFIS. The results showed that this model had high accuracy to forecast groundwater depth and was suitable for groundwater depth short-term prediction.