This paper presents a technique for segmenting spot-objects of possibly different sizes in images of biochip arrays. As these images are taken after the biochips underwent a complex chemical process, they are subject to blur effects, distortions, and local intensity shifts. Image segmentation is reduced to quantization, which in turn is reduced to function approximation. The approximation requires neither parametric assumptions nor parametric input and it iteratively optimizes an n-step histogram function by coarser step functions. Thus, a discrete “scale space” is created. The notion of so-called stable minima is used for finding good binarization thresholds. Stability issues are combined with heuristic criteria tailored to segmenting biochip images. Finally, the complete approach is applied to a series of real biochip images.