To realize the real-time, online, and accurate counting of a certain amount or quantities of seedlings  during farming, transportation and marketing, this research proposes an automatic counting system of fries based on computer vision. A tank device with steady flow velocity was designed according to Bernoulli principle in Fluid Mechanisms, enabling the seedlings and water to pass through the shooting area of the flow-through counting box more stably and constantly. Images were acquired using high speed CCD camera which had a frame rate consistent with the flow rate, and then transferred to a computer for image processing. The method of image threshold segmentation and target recognition was used, to work out the number of the seedlings that were non-overlapping in each frame image and then get the total number through adding up all the numbers. The experimental results showed that the relative error of the counting system was lower than 15%; in other words, it was of high precision. This counting system has solved the problems of target adhesion, continuous counting and double-counting, and could also be extended and applied in the seedling counting of shrimps, crabs and other biological organisms. In general, this method has the features of versatility, good feasibility and wide applicability.