The gear is a very important part in the mechanical movement,once the gear is damaged,it can cause inestimable losses to whole machinery,even affects personal safety.Therefore,it is very important to predict the remaining life of the gear in advance.With the rapidly development of deep learning，it can be used to predict the remaining useful life (RUL) of gears.Convolutional neural networks (CNN) has the characteristics of weight sharing and local perception,however,it still has some defects in processing time series.Gated recurrent unit(GRU) can deal with the problem of insufficient dependence over long distance of time series and has a simple structure.In order to have the feature of weight sharing and also solve the time series problem,the CNN-GRU model was proposed to predict the gear life.The experimental results show that the accuracy and training speed of this method are improved,and it has certain application value.