The application layer of Distributed Denial of Service (DDoS) has the characteristics of low rate,legitimate messages.The existing methods are mainly based on supervised learning which needs enough labeled samples.But the real network is composed of large amount of original unlabeled samples,it will take us a lot of time and money to get labeled samples.Aiming at solving this problem,a semi-supervised manifold regularization Web DDos detection method(semi-WDD) was proposed in this paper.Firstly,Web log was filtered into a 14 dimensional feature space according to IP address or domain name within a time window; Secondly,Laplacian Regularized Least Squares (LapRLS) algorithm which is based on semi-supervised manifold regularization was introduced to classify the data in the feature space to distinguish abnormal users and normal users.Through the experimental analysis,the method we proposed was contrasted with other methods in terms of adaptability of few labeled samples.Experimental results showed semi-supervised manifold regularization detection method has a better classification accuracy compared with other 5 mainly used methods in Web DDoS detection.Therefore,semi-WDD method has better practicability for detecting Web DDoS.