We consider a generic elementary gate sequence which is needed to implement a general quantum gate acting on n qubits-a unitary transformation with 4(n) degrees of freedom. For synthesizing the gate sequence, a method based on the so-called cosine-sine matrix decomposition is presented. The result is optimal in the number of elementary one-qubit gates, 4(n), and scales more favorably than the previously reported decompositions requiring 4(n)-2(n+1) controlled NOT gates.