We introduce a new approach to connectivity-dependent properties of diluted systems, which is based on the transfer-matrix formulation of the percolation problem. It simultaneously incorporates the connective properties reflected in non-zero matrix elements and allows one to use standard random-matrix multiplication techniques. Thus it is possible to investigate physical processes on the percolation structure with the high efficiency and precision characteristic of transfer-matrix methods, while avoiding disconnections. The method is illustrated for two-dimensional site percolation by calculating (i) the critical correlation length along the strip, and the finite-size longitudinal DC conductivity: (ii) at the percolation threshold, and (iii) very near the pure-system limit.The transfer-matrix (TM) approach to percolation was pioneered by Derrida and coworkers [1]. By analysing the possible combinations of adjacent column states made up of occupied and unoccupied sites (or bonds), and the allowed connections among the latter and to the arbitrary origin of a two-dimensional (d = 2) strip, it was possible to write the TM on the basis of such column states. The key element in this formulation was the fact that, from the very structure of the TM, repeated multiplication is tantamount to the simultaneous generation of all possible connected configurations that span the strip, each with its proper probabilistic weight. Thus the probability of connection to the origin, whose exponential decay is governed by the correlation length, is asymptotically given exactly by the largest eigenvalue of the TM. The correlation length could then be used in a phenomenological renormalisation calculation [2], which gave very accurate results for critical parameters such as the percolation threshold and correlation-length exponent ν. It was not clear, however, how one could take advantage of such a direct and elegant scheme to investigate properties other than the decay of the probability of connection to the origin. The obvious alternative, of building up successive columns by occupying (or not) each individual site independently, runs into the problem of disconnections, which is severely aggravated on a strip geometry. Up to now, the usual solution has been to generate configurations site by site, and study quantities which do not depend on keeping connectivity along the strip, e.g. the moments of the distribution of clusters [3] for percolation in d = 2 and 3. A clever way to get round the effects of disconnections for random resistor-insulator networks at percolation was introduced [4] by generating individual elements on long strips (or bars, in d = 3) with free edges. By imposing a fixed voltage drop across the strip, it was possible to invoke TM concepts in a step-by-step evaluation of the transverse conductivity, for which longitudinal disconnections of the resistor structure are irrelevant. Finally, in superconductor-resistor networks at the percolation threshold of superconducting elements, disconnections are in fact responsib...