The goal of computer-aided detection (CADe) systems is to aid radiologists in interpreting mammograms. Both independent studies and manufacturer reports indicate that current CADe systems often fail to detect breast cancers that present as spiculated lesions on mammography. We are developing a new evidence-based framework for the detection of spiculated lesions. By evidence-based we mean that we use measurements of the properties of spiculated lesions to guide the design and selection of parameters of our detection algorithm. When measurements made by human observers are used, it is important to understand their reliability. There are two sources of variability in measuring properties of spiculated lesions. The precise location of spicules must first be identified by the observer and then s/he must measure the quantity of interest (e.g., length). In this paper, we present a preliminary report of the observer variability segmenting and measuring spiculated lesions