1: \begin{abstract}

2:   We study the rate of convergence to a normal random variable of the

3:   real and imaginary parts of $\tr A_N U$, where $U$ is an $N \times

4:   N$ random unitary matrix and $A_N$ is a deterministic complex

5:   matrix.  We show that the rate of convergence is $O\left(N^{-2 +

6:       b}\right)$, with $0\le b < 1$, depending only on the asymptotic

7:   behaviour of the singular values of $A_N$; for example, if the

8:   singular values are non-degenerate, different from zero and $O(1)$

9:   as $N \to \infty$, then $b=0$.  The proof uses a Berry-Ess\'een

10:   inequality for linear combinations of eigenvalues of random unitary

11:   matrices, and so appropriate for strongly dependent random

12:   variables.

13: \end{abstract}

14: