Neurules are a kind of integrated rules integrating neurocomputing and production rules. Each neurule is represented as an adaline unit. Thus, the corresponding neurule base consists of a number of autonomous adaline units (neurules). In this paper, we present the construction process and the inference mechanism of neurules and explore their generalization capabilities. The construction process, which also implements corresponding learning algorithm, creates neurules from a given empirical data set. The inference mechanism of neurules is integrated in its nature; it combines neurocomputing with symbolic processes. It is also interactive, i.e., it interacts with the user asking him/her to provide values for some variables necessary to carry on inference. As shown via experiments, the neurules' integrated inference mechanism is more efficient than the inference mechanism used in connectionist expert systems. Furthermore, neurules generalize much better than their constituent neural component (adaline unit) and are comparable to the backpropagation neural net (BPNN).