High quality data smoothing is frequently required in the thermal analysis. Though the mathematical methods for smoothing are well known, the selection of the proper smoothing parameters cannot be based on statistical checks alone in thermal analysis. Many times a compromise must be found between the effective removal of the various experimental error components and the distortion of the curves by too strong smoothing. The following topics are discussed from a practical point of view: (i) the determination of the derivative thermogravimetric (DTG) curves at low sample masses; (ii) the elimination of the noise and flutter from the results of high-pressure TG experiments; (iii) the evaluation of the noisy, low intensity mass spectrometric signals arising from the minor volatile products of decomposition or oxidation. The performance of the methods is shown in two applications: (1) Evaluation 0.25-0.5 mg mass loss steps in high-pressure thermogravimetry; (2) Study of NOx formation by atmospheric pressure thermogravimetry-mass spectrometry during the temperature programmed combustion of 0.4 mg coal char.