In the context of secure point-to-point message transmission in networks with minimal connectivity, previous studies showed that feedbacks from the receiver to the sender can be used to reduce the requirements of network connectivity. We observe that the way how feedbacks were used in previous work does not guarantee perfect privacy to the transmitted message, when the adversary performs a Guessing Attack. In this paper, we shall describe our new Guessing Attack to some existing protocols (in fact, we are the first to point out a flaw in the protocols of Desmedt-Wang's Eurocrypt'02 paper and of Patra-Shankar-Choudhary-Srinathan-Rangan's CANS'07 paper), and propose a scheme defending against a general adversary structure. In addition, we also show how to achieve almost perfectly secure message transmission with feedbacks when perfect reliability or perfect privacy is not strictly required.