This paper deals with the analysis of otolith images for fish ageing issues. We present a new and well-founded approach for the automatic detection of the nucleus within otolith images. This topic is of key interest since the knowledge of the location of the otolith nucleus is required to achieve further processing (2D ring segmentation, age and growth estimation,...). Using morphological features within an a contrario framework, we develop a robust parameterless scheme. Its efficiency is demonstrated by an evaluation carried out for a set of several hundred of plaice otoliths.