Dear R-help, Is seems to me that R doesn't allow correct F-tests to be computed in anova.glm after estimating the dispersion parameter by deviance/df. Or does it? //Henric