Multi-user detection allows for the efficient use of bandwidth in Code Division Multiple Access (CDMA) channels through mitigation of near-far effects and multiple-access noise limitations. The decorrelating detector, developed by Lupas, is a linear multi-user detector that is asymptotically optimal in terms of near far resistance. In this paper, a simple adaptive decorrelating detector is developed by placing constraints on the set of spreading codes to be used by the active users. This adaptive detector has two modules: it first decorrelates the existing users and then determines the spreading code of a new user entering the network with the use of a training sequence. Maximum Likelihood detection is proposed for determining the new user's spreading code. The performance of this algorithm is studied by investigating the probability of error as a function of the length of the training sequence, the number of users transmitting, and the signal to noise ratio of the new user with respect to the ambient Gaussian noise.