Cross-lingual Named Entity Recognition (NER) has recently become a research hotspot because it can transfer knowledge from high-resource languages to low-resource languages, thus meeting the challenge of data scarcity in low-resource languages. Most of the current model-transfer methods rely on directly using multilingual models to represent text yet ignoring the cross-lingual word alignment information in multilingual models, and also neglecting the utilization of prior knowledge. We propose a dual contrastive learning method based on machine reading comprehension (MRC) framework, by combining Query Contrastive Learning (QCL) and Translation Word Contrastive Learning (TWCL), to mitigate above problems. Specifically, we utilize QCL to design contrastive objectives for different query templates to enhance the representation ability of prior knowledge. In addition, we utilize TWCL to help the model capture the word alignment relationship between the source language and target language via a pseudo-parallel corpus. We conducted extensive experiments on 4 different datasets and the experimental results demonstrate the effectiveness of our method.