In differential scanning calorimetry, the thermal properties of a sample are recorded as a function of temperature during a predetermined temperature scan. Sometimes, changes of such thermal properties are so fine that it could be interesting to differentiate the recorded signal as a function of time or temperature. This mathematical operation brings out these fine and broad signal variations by removing a 'non-interesting' background. Unfortunately, this also brings out the experimental peak to peak noise, preventing a clear observation. This work presents a new method in differential calorimetry that gives access directly to the temperature derivative of usual differential scanning calorimetry signals. With this method signals are recorded with the same level of noise than those measured by differential scanning calorimetry. As a matter of fact, by means of a mathematical integration, differential scanning calorimetry properties can be recovered with a greater signal to noise ratio. This method can be generalized to any other differential thermal techniques and calorimeters.