In high-resolution NMR, a network of scalar-coupled spins can be broken down into fragments which encode information about the chemical shifts and couplings that are relevant to a cross-peak multiplet in a two-dimensional correlation (''COSY'') experiment. Novel selective two-dimensional NMR correlation methods (''soft-COSY'') are described which allow one to identify fragments on the basis of their topology. These new experiments allow one to ''zoom in'' on complicated cross-peak multiplets and include ''passive spin filters'' (PSFs), consisting of selective inversion pulses applied at the frequencies of passive coupling partners, i.e. spins that are not directly observed in the cross-peak multiplets under investigation. It is demonstrated how such PSFs can be inserted at various points in two-dimensional experiments and how such experiments can be characterized in terms of topology. The topology, couplings, and shifts of a fragment may be represented by a matrix M, while a selective experiment can be represented by a matrix P. Several degrees of matching of M and P can be distinguished. The techniques, which provide a novel method of discriminating and assigning spin systems, are demonstrated on a cyclopropane derivative and the cyclic undecapeptide cyclosporin.