Given that autocorrelation tests do not perform well in the presence of heteroskedasticity and in variance-break cases, we present three modified weighted variance ratio tests of autocorrelation. The numerical results show that the proposed tests perform better for small samples. They provide a better approximation of asymptotic distributions and are more powerful when the lag length is mis-specified. The study also applies these tests to data on the daily returns of two companies listed on the Pakistan Stock Exchange.