The most significant problem of maize grain mechanical harvesting quality in China at present is the high grain breakage rate (BR).  BR is often the key characteristic that is measured to select hybrids desirable for mechanical grain harvesting.  However, conventional BR evaluation and measurement methods have challenges and limitations.  Microstructural crack parameters evaluation of maize kernel is of great importance to BR.  In this connection, X-ray computed microtomography (μ-CT) has proven to be a quite useful method for the assessment of microstructure, as it provides important microstructural parameters, such as object volume, surface, surface/volume ratio, number of closed pores, and others.  X-ray computed microtomography is a non-destructive technique that enables the reuse of samples already measured and also yields bidimensional (2D) cross-sectional images of the sample as well as volume rendering.  In this paper, six different maize hybrid genotypes are used as materials, and the BR of the maize kernels of each variety is tested in the field mechanical grain harvesting, and the BR is used as an index for evaluating the breakage resistance of the variety.  The crack characteristic parameters of kernel were detected by X-ray micro-computed tomography, and the relationship between the BR and the kernel crack characteristics was analyzed by stepwise regression analysis.  Establishing a relationship between crack characteristic parameters and BR of maize is vital for judging breakage resistance.  The results of stepwise multiple linear regression (MLR) showed that the crack characteristics of the object surface, number of closed pores, surface of closed pores, and closed porosity percent were significantly correlated to the BR of field mechanical grain harvesting, with the standard partial regression coefficients of –0.998, –0.988, –0.999, and –0.998, respectively.  The R² of this model was 0.999.  Results validation showed that the Stepwise MLR Model could well predict the BR of maize based on these four variables.