Three-dimensional geometric models of irregular food were constructed to contribute to process modelling area shedding light on geometry development aspects. Object reconstruction was performed using a cross-sectional design technique from magnetic resonance images. Boundaries of cross-section images were approximated by B-Spline curves after an image processing procedure. Then, a non-uniform rational B-Spline based model representing the food material was built using a lofting method by interpolating all B-Spline curves. This methodology was applied to lamb, pork and chicken carcasses, and beef semitendinosus muscle. The obtained geometric models were visually compared with real samples showing a very good agreement. The presented methodology can be used to generate a geometry database saving efforts and decreasing error associated to experimental measurements. (c) 2008 Elsevier Ltd. All rights reserved.