The size of the Jarque–Bera test for multivariate normality can be severely distorted in small samples. An alternative test procedure, that turns out to have good size and power properties, is suggested.