Generation of covering arrays, which has been solved by many mathematical methods and greedy algorithms as well as search based algorithms, is one of significant problems in combinatorial testing. As an effective evolutionary search algorithm for solving combinatorial optimization problems, ant colony optimization has also been used to generate covering arrays. Existing research shows ant colony optimization suitable for generating general covering arrays, variable strength covering arrays and the prioritization of covering arrays. Unfortunately, compared with other methods, ant colony optimization doesn't have significant advantages. To further explore and mine the potential of ant colony optimization in generating covering arrays, this paper focuses on four levels of improvement:1) the integration of ant colony variants; 2) parameter tuning; 3) the adjustment of solution structure and the improvement of evolutionary strategy; 4) using parallel computing to save executing time. The experimental results show that ant colony optimization is much more effective in generating covering arrays after the improvements.