A time-redundant approach is proposed by adopting a rollback and/or roll-forward technique to recover TMR failures producing incorrect majority outputs in a TMR (structured) system that uses the simplest spatial redundancy. This technique is apparently effective in recovering TMR failures primarily caused by transient faults. The proposed policies carry out fewer reconfigurations at the cost of (minimal) time overhead needed for applying those time-redundant schemes. Specifically, two methods are presented (flexibly used) for the recovery policy to guarantee high reliability. The optimal checkpoint vectors are also derived for both methods by using the likelihoods of all possible failure states as well as the total execution time. Consequently, the effectiveness of the proposed policy is validated through certain numerical examples and computer simulation.