Noise which occurs in space and time temperature evolution is analyzed using three informative parameters: temperature signal over defect, running contrast, and normalized contrast. It is shown that the choice of optimum informative parameter depends on the kind of noise which dominates in a particular experiment. Decrease of noise by passing into time domain is demonstrated. A methodology for creating images of signal-to-noise ratio and probability of correct detection is proposed. Experimental validations were performed for plaster specimens. The proposed methodology allows us to compare various thermal NDT techniques.