Student engagement, classes with learning activities, and students' soft skill development are a few desired characteristics of a successful class. However, these are difficult to achieve for courses that deal mostly with theory and simulation. This paper presents the first experience in implementing a desktop project to enhance the student learning of such a course. "Artificial Neural Network" course at College of Engineering Technology, Can Tho University, was chosen to evaluate the proposed approach since its content focused largely on the theory of multilayer feedforward networks and simulation of these networks to solve fitting or pattern recognition problems. The paper helicopter project, well known for teaching statistics, was chosen as the desktop project in the course as a fitting problem, i.e., finding the landing time of the paper helicopter from a certain height. This was chosen because it was easy to implement and students could work in teams in a real experimental design. Within three hours' time for the desktop project, students could design the experiment, gather and process the data, develop and evaluate an artificial network to give the helicopter's landing time from its height. After the class, students could perform the experiment again to obtain more data, retrain the network for better performance so that they could hand in a good report. It was apparent that students engaged more in the class and they developed more soft skills in presenting their work. Students' feedback from a survey showed that all students enjoyed the project and even request more time for similar activities. From the results of the survey and the end-course survey, some lessons were learned and recommendations for similar courses are proposed.