Cross domain object detection is challenging, because object detection model is often vulnerable to data variance, especially to the considerable domain shift in cross domain scenarios. In this paper, we propose a new approach called Unbiased Mean Teacher (UMT) for cross domain object detection. While the simple mean teacher (MT) model exhibits good robustness to small data variance, it can also become easily biased in cross domain scenarios. We thus improve it with several simple yet highly effective strategies. In particular, we firstly propose a novel cross domain distillation for MT to maximally exploit the expertise of the teacher model. Then, we further alleviate the bias in the student model by augmenting training samples with pixel-level adaptation. The feature level adversarial training is also incorporated to learn domain-invariant representation. Those strategies can be implemented easily into MT and leads to our unbiased MT model. Our model surpasses the existing state-of-the-art models in large margins on benchmark datasets, which demonstrates the effectiveness of our approach.