This paper describes a strategy of optimal action of sensing process. The decision of an action selection is made from an information theoretic point of view. Its goal is to reduce the number and effort of the sensing steps to reach the final identification of the object condition among those registered in the data base. We have chosen an information theoretic framework, motivated by the fact that sensored data is not noiseless or ideal, nor can the effect of a certain action be completely determined in advance. This is the common case for almost all sensing problems. We present the strategy for an example of the object recognition process from observations of its images. In the recognition, the entropy, which can be considered as the amount of the uncertainty of the object, must be most rapidly reduced. This process is formalized in terms of evaluation of the mutual entropy.