Traditionally, monitoring solutions are based on collecting a reduced set of external metrics about the system such as performance, memory consumption or response time. However, these tools are limited to detecting and diagnosing failures or errors in complex systems like application servers and Grid services. New applications have embedded monitoring logic merged with business logic to better monitor and trace applications. This approach creates dependence between the ”real” code and the monitoring code though, reducing the monitoring approach flexibility. Furthermore, both approaches (external or embedded monitoring) are not adaptable; they cannot change monitoring process level or precision at runtime. In this paper, we present a fine-grain monitoring framework architecture based on aspect-oriented technology. Aspect Oriented Programming offers the possibility to inject monitoring code dynamically, and even activate or deactivate monitoring code at runtime, allowing the collecting of metrics at the appropriate granularity (i.e. application level or method calls level).