Background. This work focuses on problems of heuristic techniques based on artificial intelligence. Mainly about artificial non-linear and multilayer neurons, which were used to estimate the bone union fractures treatment process using orthopaedic stabilizers Dynastab DK. Material and methods. The author utilizes computer software based on multilayer neuronal network systems, which allows to predict the curve of the bone union at early stages of therapy. The training of the neural net has been made on fifty six cases of bone fracture which has been cured by the Dynastab stabilizers DK. Using such trained net, seventeen fractures of long bones shafts were being examined on strength and prediction of the bone union as well. Results. Analyzing results, it should be underlined that mechanical properties of the bone union in the slot of fracture are changing in nonlinear way in function of time. Especially, major changes were observed during the forth month of the fracture treatment. There is strong correlation between measure number two and measure number six. Measure number two is more strict and in the matter of fact it refers to flexion, as well as the measure number six, to compression of the bone in the fracture slot. Conclusions. Consequently, deflection loads are especially hazardous for healing bone. The very strong correlation between real curves and predicted curves shows the correctness of the neuronal model.