Ultrasonic testing is widely used in material inspection. Forming an u
ltrasonic image helps the operator in decision making. This image, how
ever, can result in common interpretation difficulties during the test
ing of heterogeneous or noisy zones. A new method of detection of defe
cts located in a noisy medium is presented: it is based on considering
the ultrasonic grey-level image from the beginning of the analysis. T
he data processing used here is searching for a certain determinism in
the spatial and temporal evolution of the image in the presence of a
defect: a first criterion studies the horizontal stability of the grad
ients in the image and a second considers the transient temporal natur
e of the defect echo. Results obtained on artificial defects located i
n real welds are shown. Conclusions with respect to the quality of thi
s defect detection method are drawn.