The preoperative assessement by magnetic resonance imaging (MRI) of mesorectum involvement could improve the treatment strategy for patients with rectal cancer. However, the anatomical definition of the mesorectum remains controversial and consequently the accurracy of its analysis by preoperative imaging workup is still unsatisfactory. The aims of this study were to define the mesorectum anatomically and to assess whether it could be evaluated accurately by MRI. Total mesorectal excision was performed in 37 patients with rectal cancer. The mesorectum was inked for anatomical analysis, which was performed before and after fixation in formalin. The mesorectal thickness was measured anteriorly, posteriorly and laterally. Mesorectal involvement was defined by the shortest distance from the outermost part of the tumour to the lateral mesorectal margin. The anatomical measures were compared to those evaluated by preoperative MRI. In middle rectum, the anatomical analysis showed that the maximal mesorectal thickness was 60 and 20 mm posteriorly and anteriorly, respectively. In low rectum, the mesorectum was very thin and its maximal thickness was less than 10 mm anteriorly and posteriorly in all cases. The mesorectal involvement was less than 2 mm in 23% of cases. In terms of mesorectal involvement, there was good agreement between anatomical analysis and MRI for middle rectum. In contrast, the agreement was fair for low rectum. This anatomical analysis could explain the poor performance of MRI in the assessement of mesorectum involvement in low rectum.