The development of next-generation high-throughput DNA sequencing techniques has greatly promoted the research of structural variations (SVs) detection.Current genetic structure variation detection methods are mainly base on depth of coverage, pair-end mapping clusters, or sequence assembly, some of them are known to be not accurate or too sensitive.What's more, some methods are not able to recognize the specific position and sequence of structural variation.Insertions and deletions (indels) are the most common forms of genome structure variations.This paper puts forward an optimal split-read matching algorithm (OSRM) using dynamic programming.OSRM breaks an abnormal read into several reads in a least quantity.First, a score matrix of the abnormal read and the corresponding referenced sequence is created.Then a matrix of backtracking path is established.Next, a formula designed according to the characteristics of structural variation is used to elect the optimal backtracking path matrix.And finally the split-read and referenced sequence are matched in an optimal arrangement by which the accurate position and sequence of found indels are outputted.Experiments prove that the performance of algorithm is excellent.In addition, compared with Pindel which is the best in split-read methods, OSRM can offset its defection in detecting small and medium indels while also be able to detect more complex situation.