We propose a methodology for observer synthesis for a class of chemical reactions processes. First we present graphic tools in order to decide of the property of observability of the model of the reaction process and/or the best measures to do and which kind of observer to use (high gain observers in the mono output case and Kalman observers). Next we give an attractivity definition. Thanks to these graphic tools and this attractivity property valid for some part of CSTR's, we obtain in many cases, the estimation of the whole state even if some part of the model is not observable. For this purpose, we use cascades of the previous observers and estimators. Moreover we present the methodology via algorithms and we give a simulation example.