The invention discloses a maximal connected subgraph-based human face identification data cleaning method. The method sequentially comprises the following steps of dividing collected photos into multiple photo groups according to tags; comparing each photo in each photo group with other photos in the photo group one by one to obtain multiple in-group photo similarity values; constructing an in-group similarity graph of each photo group according to the in-group photo similarity values; finding out a maximal connected subgraph of each photo group; deleting the photos not belonging to the maximal connected subgraphs in all the photo groups; taking the photo corresponding to a node with a maximum edge number in each photo group as a reference photo; and comparing the reference photo of each photo group with reference photos of other photo groups one by one to obtain inter-group photo similarity values, and combining similar photo groups. According to the method, the operation of cleaning collected photo data is low in cost and small in workload, and wrong photos in the photo data can be effectively deleted.