New technologies, such as three-dimensional (3D) laser scanning and stereo imaging, have recently been adopted for quantifying plant structure. The datasets collected using such technologies offer realistic representations of the morphological characteristics of the studied plant organs. The datasets, however, are very large and occupy excessive amount of storage space. Moreover, the computation time is also very long when these datasets are made the subject of further analysis and simulation. Some dataset simplification is essential if the balance between storage cost and computation time vs the accuracy of the plant geometry description is to be optimised. In this study, the surface morphologies of field-grown maize and tobacco leaves were measured using 3D laser scanning and were progressively simplified using two different methods – Vertex removal and Edge collapse. To evaluate the impacts of simplification on the accuracy of the leaf-surface morphological descriptions, several error metrics were developed. These metrics are able to quantify these impacts in various respects. The statistical results show that most error metrics increase only marginally, even with moderate simplifications of the leaf surfaces. The errors, however, increase quickly with over-simplification. The simulation results of light distribution in canopies indicate that over-simplification of the leaf-surface meshes results in significant deviations in the simulated leaf light-capture efficiency compared with the original leaf surfaces. Compared with the Vertex removal method, the Edge collapse method is better at retaining the original leaf-surface morphology, but loses more of the leaf-edge information. This study provides valuable information in the analysis of high-precision plant-structure data, relevant in a range of research fields including functional–structural plant modelling and high-throughput phenotyping.