Abstract: Modeling and simulation of agricultural products during harvesting, storage, and transportation process are very important issues in agricultural postharvest engineering field. Considering transport phenomena principles involved in the model formulation, a geometry representing the real object is needed to be defined. However, since agricultural products are usually naturally irregular in shape, accurate acquisition of their geometric model becomes essential. In this research, the model reconstruction process which consists of magnetic resonance imaging, image processing and reconstruction method allows obtaining accurate geometric models even for agricultural products with very irregular shapes. Some typical agricultural products with irregular shapes, such as mango, banana and apple, were selected as the experimental samples to reconstruct their geometry models from magnetic resonance imaging data. All NMR data were acquired from a GE whole-body magnetic resonance imaging system with a static magnetic field of 3.0 Tesla. Samples were fixed on a plastic holder and then manually focused to obtain the images. MR images were generated using a fast spin echo sequence with effective echo time of 65 ms, repetition time of 1800 ms, slice thickness of 2 mm, and field of view of 512512 pixels. In order to better acquire the irregular contour curves of each MR image, several image preprocessing steps implemented in MATLAB were applied to all original the MR images, the steps are as follows: (1) gray level transformation of the original gray level images; (2) median filtering for noise reduction and image quality enhancement; (3) segmentation through an adaptive threshold segmentation algorithm; (4) image filling and ensure each image only has a single contour curve. A procedure to obtain geometric models of irregular agricultural products was developed by the computer image graphics technology. The whole model reconstruction process was implemented in Visualization Toolkit 5.4. The contour curves approximating the real boundaries of cross section were correctly assembled by means of a reconstruction technique Marching Cubes algorithm. And then Laplacian mesh smoothing and Quadric Error Metric (QEM) edge folding simplified algorithm were used to enhance the practicality of the reconstruction model. In order to verify the reliability of the reconstructed methods, the obtained geometric models were converted to STL format, and then transformed to a 3D solid object. Then we imported them into a computer aided design software Pro/E to calculate their geometric parameters. Similarly, actual geometric parameters of the samples were measured by a vernier caliper and the water displacement method. Comparing the geometric model of agricultural products with the physical object, it was found that the obtained geometric models were well consistent with real samples, they showed very good agreement in shape, volume and other morphological parameters, and the errors between them were less than 3%. The presented methodology can avoid making great efforts in experimental measurements and consequently development of the geometric models, decreasing error generated from manual data extraction. The method has the significant advantages of producing geometric models of agricultural products that are easy to use in computer aided design software for numerical simulation, finite element analysis and quality detection, etc. This work is expected to be a useful contribution for modeling and simulation of agricultural products in postharvest engineering processes.