In this work, we present a conditional model for online handwriting recognition. Our approach is based on Conditional Random Fields (CRFs), a probabilistic discriminant model that has been generally used up to now in particular settings, for labeling and parsing of sequential data such as text documents and biological sequences. We propose to adapt these models in order to build systems for handwriting recognition. We propose a few systems whose architecture allows dealing with multimodal classes and exploiting segmental features that are much adapted to signal data like online handwriting.