We present an ongoing experience aimed at introducing statistical process control techniques to one crucial test phase, namely Function Test, of a real world software development process. We have developed a prediction procedure, using which, among other things, we compare the performance of a Classical model vs. a Bayesian approach. We provide here the description of the prediction procedure, and a few examples of use of the models over real sets of data. However, far from aimed at identifying new statistical models, the focus of this work is rather about putting measurement in practice, and in easy and effective steps to improve the status of control over test processes in a soft, bottom-up approach. The experience described has started quite recently, and the results obtained so far, although limited in scope, are quite encouraging (as well as exciting for the involved team), both in terms of predictive validity of the models and of the positive response got from development personnel.