Abstract: Abstract: Agricultural soil cracks generated during drying process becomes one of the causes of preferential flow. Generalizing crack network structure reasonably is an important research in the field. A laboratory experiment was carried out to investigate the law behind the development of shrinkage cracks on the soil surface under different types of cultivated land. In this investigation, computer image processing technique was applied to quantificationally analyze the statistical law of crack morphology and a dynamic description method based on normal distribution was proposed to depict the morphology of shrinkage cracks on soil surface. It showed that crack area density and length density both increased with decreasing moisture content and remained stable after reaching their maximums. Crack area density of paddy field (PF) specimen was a little bit bigger than that of upland field (UF) specimen throughout the drying process while the crack length density of both specimens didn't appear to be much different under the same mass moisture content. The connectivity index increased with decreasing moisture content and finally tended to be stable. The maximum connectivity index of PF specimen was 0.935 around a mass moisture content of 20% while for the UF specimen was 0.598 around a mass moisture content of 12%, reflecting that cracks of PF specimen were more closely connective to each other and the crack network of PF specimen formed faster. The connectivity index is applicable to account for the forming process and actual connectivity of shrinkage crack network under different types of cultivated land. Results suggested that crack area probability density and cumulative probability were in line with normal distribution. Both mean value and standard deviation had fairly good correlation with moisture content. Their coefficients of determination of UF specimen were larger than 0.99 and 0.98, respectively. The crack area coefficient of determination of different farmland types between the measured values and theoretical values of probability density function were both larger than 0.94, while for the crack length were both larger than 0.91, and their root mean square error(RMSE) were both less 0.20. The crack area coefficient of determination of different farmland types between the measured values and theoretical values of the distribution function were both larger than 0.90, while for the crack length were both larger than 0.94. And their RMSE were both less than 0.12. This research indicated that normal distribution is appropriate to depict crack morphology on soil surface through, which is feasible to study the formation mechanism of soil crack and the effect of different crack shapes on soil preferential flow.