Abstract: After defining a effective index for process similarity,an analogue model,which is suitable for doing similarity forecast fora time series,was established.The model was applied to an importantindex for ENSO's monitoring-NIN O3 SST anomaly,and some results had been discovered.Firstly,similarity forecastis obviously better than persistence forecastin prediction effects such as mean absolute error and correlation co efficient,especially in 6-8 months lead predictions,the maximum lead of effective prediction is 8 months,so the prediction ability of the model almostly reaches the level of the models in the world.Secondly,for the effective lead time(≤8 months) predictions,the correlation skills are generally better in winter half year than in summer half year,with the best in Decemberand the worst in June and July.M eanwhile,forall lead predictions,the smallest mean absolute erroroccurs during February to April,and the larggest in December.Finally,similarity forecast has better ability in turn-point prediction,especially for El Nino's ending and strong El Nino's beginning.