Public response to the pandemy since March 2020, including lockdowns, has led to the massive use of remote evaluation, in the form of online tests, of the MCQ type or related (questions with numerical answers, graphics, calculated, etc.). Since these tests are carried out without supervision, it is tempting for some students to exchange information during the test. This exchange of information, when it takes place, leaves a mark in the form of statistical anomalies in the answers to the questions as well as in the timestamp. The objective of our work was to propose and test methods aimed at automatically detecting these anomalies, in order to draw the teacher's attention to suspicious cases. In our field (electronics, digital communications, image and signal processing, ...) the small number of students (a few dozen) and questions makes complex parametric models, such as those found in the literature, unsuitable, the latter being designed for tests at very large scale (with thousands of candidates) providing enough data to fit the model. We describe the developed method and we illustrate it through the actual results obtained on three on-line tests.