Luus’ random search algorithm does not converge to the optimal solution when the selected initialvalue X~((0)) and initial search region R~((0)) are not proper.This paper has found out a reason of thefact that the above mentioned algorithm took a systematic reduction of the sizo of search region underthe constant factor.The modified random search algorithm has been proposed by the author.The idea of this al-gorithm lies in that the search region is expanded when the search is proceeding with big strides,viceversa.Thus the search region was reduced from the overall situation,but,on the contrary,it wasexpanded occasionally from the partial situation.The modified algorithm is noted as below:whether X~((0)) or R~((0)) was selected has no influence overthe reliability;it goes a way towards solving the problem of rather high dimension in theory;it issuitable not only to solving the problem of constrained nonlinear programming but also to that of un-constrained;the feasible region is not necessarily limited on convex region.The application of the modified random search algorithm was described with illustrations and thecomparison between modified and Luns’ random search algorithm has been made. Luus’ random search algorithm does not converge to the optimal solution when the selected initial value X ~ ((0)) and initial search region R ~ ((0)) are not proper. This paper has found out a reason of the factor that the above mentioned algorithm took a systematic reduction of the sizo of search region underthe constant factor. The modified random search algorithm has been proposed by the author. The idea of ​​this al-gorithm lies in that the search region is expanded when the search is proceeding with big strides, viceversa.Thus the search region was reduced from the overall situation, but, on the contrary, it wasexpanded occasionally from the partial situation. The modified algorithm is noted as below: whether X ~ ((0)) or R ~ (( 0)) was selected has no influence overthearity; it goes a way towards solving the problem of rather high dimension in theory; it issuitable not only to solve the problem of constrained nonlinear programming but also to that of un-constrained; the feasible region is not nece ssarily limited on convex region. The application of the modified random search algorithm was described with illustrations and the comparison between modified and Luns’ random search algorithm has been made.