The automated detection of cell nuclei, which is an important step in the pipeline of quantitative histopathological analysis, has received considerable attentions in recent years. However, biological variations, uneven staining and illumination, non-rigid deformations and touching or overlapping of the cell nuclei have made the detection procedure a major hurdle. In this paper, we consider the problem of detecting and localizing cell nuclei based only on the cue of contour. The detectors are learned with a boosting algorithm which creates a pixel classifier using an over-complete set of contour features with different curvatures and orientations. We present results that are very competitive with other state-of-art cell nuclei detection scheme and show robustness to biological variations, different staining and illumination conditions, touching or partial occlusions. Our major contributions are the multicurvature cell nucleus contour model, and an efficient and effective strategy for sharing parameters between contour feature extractors of the same curvature but different orientations.