We propose a reference-free method for identifying potential classification errors and quantitatively evaluating the reliability of vector-based land cover data. First, by integrating the results of multiple-resolution segmentation on a data-driven scale, land cover data are divided into segments with increased homogeneity for feature extraction. Then, the image features of each land class are extracted and outlying segments are identified through an adaptive clustering process. Last, an entropy-based reliability measure (EBRM) is proposed to estimate the reliability of land patches. In EBRM-based evaluation, the diversity and abnormality of segments within a land patch are considered. The proposed approach is compared with state-of-the-art methods on two real-life data sets and shows better performance in terms of error identification. The effectiveness of the quantitative description provided by EBRM is validated in the experiments.