Tests for heteroskedasticity in the classical linear regression framework typically require the researcher to specify the form of variance heterogeneity. This paper proposes a simulation-based test with power to detect unspecified forms of heteroskedasticity.