In this paper, we investigate the model-driven deep learning (DL) for joint MIMO channel estimation and signal detection (JCESD), where signal detection considers channel estimation error and channel statistics while channel estimation is refined by detected data and takes the signal detection error into consideration. In particular, the MIMO signal detector is specially designed by unfolding an iterative algorithm and adding some trainable parameters. Since the number of trainable parameters is much fewer than the data-driven DL based signal detector, the model-driven DL based MIMO signal detector can be rapidly trained with a much smaller data set. Furthermore, the proposed signal detector can be extended to soft-input soft-output detection easily. Based on numerical results, the model-driven DL based JCESD scheme significantly improves the performance of the corresponding traditional iterative detector and the signal detector exhibits superior robustness to signal-to-noise ratio (SNR) and channel correlation mismatches.