When loop transfer recovery (LTR) is considered to be a special case of sensitivity shaping, various design methods emerge, some of which relax the requirement that the estimator be unbiased. Each of these methods emphasizes a certain frequency range in the recovery process, hence providing the designer with more options for sensitivity shaping. In this paper various properties of these methods are analyzed. It is shown that the methods can be reformulated in order to make use of recent results in H-infinity optimization. Hence, although the recovery problem is formulated in the intuitively appealing frequency domain, the computations are carried out in the computationally robust time domain. As the H-infinity cost is allowed to approach infinity, the design procedure converges to the LQG/LTR procedure. In fact, many published optimization approaches to LTR can be shown to be special cases of the general methodology of this paper. The relationships between the quality of the full state feedback, nonminimum-phase zero location and direction, and the amount of recovery achievable with these methods are also analyzed. The design procedure can also be used to reduce sensitivity to parameter uncertainties. This allows one to trade off the sensitivity to structured and unstructured uncertainties within the same framework.