In this paper, we propose an algorithm for detecting the number $M$ of Gaussian sources received by an array of a number $N$ ($N< M$ ) of sensors. This algorithm is based on the minimum description length (MDL) principle using the outer-products of the array output. We show that as long as the covariance matrix of the array output has the full rank $N$, the covariance matrix of a vectorized outer-product of the array output has the full rank $N$-squared, which meets a validity condition of the MDL algorithm. We show by simulation that the MDL algorithm can perform substantially better than some relevant algorithms. A necessary identifiability condition is also obtained, for uncorrelated sources.