Applying Randomized Hough Transform (RHT) directly to the detection of ellipses, five points were randomly sampled to determine the ellipse parameters, which produces a large amount of useless accumulations when processing complex images. To solve the problem, a new ellipse detection method using RHT was proposed, which uses two randomly picked points (with edge directions) and one searched point to calculate the ellipse parameters. In this method, only two points are randomly picked when calculating the ellipse parameters, furthermore, the searched point can be used to determine whether the parameters should be accumulated or not, therefore, the problem of useless sampling and accumulations is well solved. In comparison with other detection methods, this method has the advantages of high speed, small storage and high detection performance.