EEG decoding systems based on deep neural networks have been widely used in decision making of brain computer interfaces (BCI). Their predictions, however, can be unreliable given the significant variance and noise in EEG signals. Previous works on EEG analysis mainly focus on the exploration of noise pattern in the source signal, while the uncertainty during the decoding process is largely unexplored. Automatically detecting and quantifying such decoding uncertainty is important for BCI motor imagery applications such as robotic arm control etc. In this work, we proposed an uncertainty estimation model (UE-EEG) to explore the uncertainty during the EEG decoding process, which considers both the uncertainty in the input signal and the uncertainty in the model. The model utilized dropout oriented method for model uncertainty estimation, and Bayesian neural network is adopted for modeling the uncertainty of input data. The model can be integrated into current widely used deep learning classifiers without change of architecture. We performed extensive experiments for uncertainty estimation in both intra-subject EEG decoding and cross-subject EEG decoding on two public motor imagery datasets, where the proposed model achieves significant improvement on the quality of estimated uncertainty and demonstrates the proposed UE-EEG is a useful tool for BCI applications.