Agricultural changes have caused severe decline in the common bird communities in Europe. Mitigating this loss requires both understanding of and way to predict how agriculture affects biodiversity. The objective of this paper is to test the different dynamic models coupling bird abundances with farming land use. The agro-ecological calibration relies on 2002-2009 data for 34 bird species and 14 agricultural systems in 620 small agricultural regions of metropolitan France. The models are compared based on indicators of goodness-of-fit and predictive quality. The results highlight the relevance of systemic, functional, and mechanistic relationships between the agriculture and bird populations in both descriptive and predictive contexts. However, it seems that the best dynamic models to describe the past trends are not necessarily the most relevant to predict the future trends although there is a similar subset of five models that emerges in both cases. Ranking these different models depends on both the objective (describing or predicting) and the chosen functional level for the analysis (total community, functional groups or species).