This study proposes a multiresolution Markov random field model with fuzzy constraint in wavelet domain (MRMRF-F). In this model, a fuzzy field is introduced into the multiresolution Markov random field model to estimate parameters, by which the spatial constraint between neighbouring features can be reflected. There are three subfields on each resolution in the MRMRF-F model: one feature field, one label field and one fuzzy field. Among these fields, a three-step iteration scheme is designed to realise image segmentation. Namely, the label field renews the fuzzy field; then the fuzzy field estimates the parameters of the feature field; and a renewed label field is obtained by maximising the products of both the probability of the feature field and the probability of current label field; image segmentation is finally obtained by the cycle iteration among these fields. Texture and natural images are used to test the performance of the new model, and experimental results illustrate a better and higher accuracy compared with other existing methods about fuzzy methods or Markov random field models.