We compare two approaches to Semi-Riemannian metrics of low regularity. The maximally reasonable distributional setting of Geroch and Traschen is shown to be consistently contained in the more general setting of nonlinear distributional geometry in the sense of Colombeau.