Height growth of Caragana korshinskii in semi-arid loess hilly region was not always monotonous increasing with time. Because growth power was less than growth resistance in late growth stage, plant height weakly reduced with time. In this article, two growth model were established. Adding a growth resistance factor to the classical Logistic model, the first one was improved growth model, which made growth rate negative in late growth stage. The second one was multi-year growth model, which was based on the growth observation data of ripe Caragana korshinskii with a high density of 87 brush per 100 square. The result showed that the improved model had higher goodness of fit and correlation coefficient than that of Logistic model using modeling data to fit and test data to verify. Model established in this article was different with classic plant growth model. Because resistance factor was introduced into differential equation, growth curve, where extreme coordinate exist, was not strict monotonous. Because multi year height growth curves were synthesized into one coordinate system, the ratio of position parameter a and intrinsic growth rate b in new model became more and more large with growth. Growth vertex appeared in middle ten days of August, which was consistent with the actual growth process of height. The difference of growth vertex and plant height in late growth stage, which was called growth loss, was calculated. Because Caragana korshinskii shrub was low, the process that dry topped phenomenon of Caragana korshinskii could influence on plant height in winter could not be ignored. This process caused the height at initial growth point in second year lower than that at last growth point in first year. In consideration of dry topped phenomenon, termination point of the winter drying phenomenon in multi-year growth model from 2002 to 2003 located in February, which was consistent with the rhythm of plant growth. This research provided evidence for describing growth process of shrub in semi-arid region.