In this report, I present a peak detection method for 1D data, based on scale-space theory. Instead of focusing on local derivative information as is classical in peak detection, the proposed approach is more global. It performs iterative smoothings of the input data with increasing length-scales and then defines a peak as a datapoint that remains a local maximum for many such filterings. Formally, the local maxima are identified after each filtering operation and then associated to the maxima identified with the previous length-scales. A score is then added to the criterion for these latter points, that notably depends on the length-scale. This strategy enforces picks that remain local maxima even after many smoothing operations. At the end of the process, the peaks are identified as the points having the largest score. The approach is flexible enough to allow for different smoothing operations and different strategies for incrementing the score. I informally show on different kinds of signals that the proposed approach may be very effective, even for very noisy data.