Some transformation operations for logic programs, basic for partial deduction, program specialization, and transformation, and for program synthesis from specifications, are studied with respect to the minimal S-model semantics defined in [31, 15–17]. Such a semantics is, in our opinion, more interesting than the usual least Herbrand model one since it captures the program's behavior with respect to computed answers. The S-semantics is also the strongest semantics which is maintained by unrestricted unfolding [31]. For such operations, we single out general applicability conditions, and prove that they guarantee that the minimal S-model semantics of a program is not modified by the transformation. Some sufficient conditions, which are very common in practice and easy to verify, since they are mostly syntactical, are also supplied with simple exemplifications.