This paper studies the feasibility of applying machine vision technique to minnows-counting. The content of the present study is focused on the exploration of image processing method. Also, the cost-effective machine vision system associated with programming a high efficient software of image processing is developed to meet the requirement of on-line operation. The steps of image processing methods for accurately counting minnows is: Freezing image→Binarying image→Opening operator→Closing operator→Counting minnows by recursive method. The results of testing indicate that the accuracy of counting minnows approaches above 99% as the numbers of minnows in shallow bowl is less than 70 numbers. The overall computational time of image processing for the present program requires about 3 seconds.