A method includes, at a resistive memory device, determining an average effective reference resistance level based on a first effective reference resistance and a second effective reference resistance. The first effective reference resistance is based on a first set of reference cells of the resistive memory device and the second effective reference resistance is based on a second set of reference cells of the resistive memory device. The method includes trimming a reference resistance at least partially based on the average effective reference resistance level. Trimming the reference resistance includes, in response to determining that the first effective reference resistance is not substantially equal to the average effective reference resistance level, modifying one or more states of one or more magnetic tunnel junction devices associated with the first effective reference resistance.