Using a character expansion method, we calculate exactly the eigenvalue density of random matrices of the form M † M where M is a complex matrix drawn from a normalized distribution P (M) ∼ exp(−Tr AMBM † ) with A and B positive definite (square) matrices of arbitrary dimensions. Such so-called "correlated Wishart matrices" occur in many fields ranging from information theory to multivariate analysis.Physicists usually think of Wigner and Dyson as the fathers of random matrix theory [1,2]. However, twenty years before their first work on the subject, Wishart [3] examined random matrices of the form MM † as a tool for studying multivariate data. The properties of these so-called Wishart Matrices, which are viewed as "fundamental to multivariate statistical analysis" [4], also find important applications in fields from information theory and communication [5][6][7], to mesoscopics [8], to high energy physics [9], to econo-physics [10].In many cases one is interested in Wishart matrices where the elements of M are not completely independent random variables, but have correlations along rows and/or columns. Important examples of this case occur in data analysis problems [11], where random samples have temporal and spatial correlations, and particularly in wireless communication and information theory [5,6]. The purpose of this paper is to derive the eigenvalue density of correlated complex Wishart matrices exactly. A forthcoming longer paper will give more details of the derivation as well as discussing certain applications in depth. This problem has previously been studied in the limit of large matrices where perturbative expansions in 1/N can be quite effective [11,6]. If either A or B is proportional to unity, simpler techniques can be used [12].We first define the problem more precisely. Let M be an N by N ′ complex matrix chosen from a normalized distributionwith A and B positive definite square matrices which define the correlations, and Tr indicates trace. Here,N and the factors of π, are normalization constants. An equivalent definition would be to let M = A −1/2 Z B −1/2 where Z is a random complex matrix with independent entries of zero mean and unit covariance. Note that A is N by N and B is N ′ by N ′ . Without loss of generality, we can assume N ≥ N ′ . For any operator O(M) we define the expectation bracket O to be an average over realizations of M so thatNote that the normalization is such that 1 = 1.Let λ n be the N ′ eigenvalues of M † M or equivalently the N ′ nonzero eigenvalues of MM † (we will also have N − N ′ eigenvalues of MM † precisely zero). We define the following quantities to calculate:where θ is the step function, λ is assumed real, and in going from Eq. 4 to 5 we have used lim ǫ→0 Im log(−y+iǫ) = πθ(y) which is true for real y. The quantity we are most interested in is the eigenvalue density ρ(λ). From Eqs. 2-6 it is clear that we can obtain ρ by calculating G ν (z). Below, we will showQ ν (z) −1 = ∆ N (a)∆ N ′ (b)(−z) N ′ (N ′ −1)/2 J ν (8)