In the multi-ellipse detection especially when they are overlapped,the existing methods such as the Randomized Hough Transform,show drawbacks of poor stability and low accuracy. In this paper,a new methodology was proposed based on the connected convex curve searching and ellipse fitting. Firstly,the edge curve is clustered into different groups by the convexity—leading to each group belonging to a certain ellipse. Secondly,the ellipse fitting is performed on grouped curves. Finally,curves on the same ellipse are merged. Since our method can make full use of properties of ellipse curves,it works well in experiments of both synthetic images and real-word images with a higher accuracy and a certain anti-noise performance.