In this paper, we propose a new two-step algorithm (PDTA) to solve the problem of underdetermined blind separation, where the number of sensors is less than that of source signals. Unlike the usual two-step algorithm, our algorithm’s first step is to estimate the number of source signals and the mixture matrix instead of K-mean clustering algorithm, in which people often suppose that the number of source signals is known when they estimate the mixture matrix. After the mixture matrix is estimated by PDTA, the short path algorithm is used to recover source signals. The last simulations show the good performance of estimation the number of source signals and recovering source signals.