We develop an approximate analysis of an arbitrary size homogeneous unreliable transfer line under a CONWIP discipline. Mean total work in process computation flows from the premise that the transfer line taken as a whole behaves essentially like a Kanban controlled macro cell with its input/ouput behavior dictated respectively by the reliability statistics of the first machine, and the probability of availability of work in process at the end of the CONWIP section. The latter is obtained from the iterative solution of a system of weakly coupled linear differential equations with a number of coupling unknowns equal to the square of the CONWIP loop length. Theory based results appear fairly accurate when compared to Monte Carlo simulation results.