Background: The goal of the brain is to provide right on time a suitable earlier-acquired model for the future behavior. How a complex structure of neuronal activity underlying a suitable model is selected or fixated is not well understood. Here we propose the integrated information \mathrm{\Phi } as a possible metric for such complexity of neuronal groups. It quantifies the degree of information integration between different parts of the brain and is lowered when there is a lack of connectivity between different subsets in a system. Methods: We calculated integrated information coefficient (\mathrm{\Phi }) for activity of hippocampal and amygdala neurons in rats during acquisition of two tasks: spatial task followed by spatial aversive task. An Autoregressive \mathrm{\Phi } algorithm was used for time-series spike data. Results: We showed that integrated information coefficient \mathrm{\Phi } is positively correlated with a metric of learning success (a relative number of rewards). \mathrm{\Phi } for hippocampal neurons was positively correlated with \mathrm{\Phi } for amygdalar neurons during the learning requiring the cooperative work of hippocampus and amygdala. Conclusions: This result suggests that integrated information coefficient \mathrm{\Phi } may be used as a prediction tool for the suitable level of complexity of neuronal activity and the future success in learning and adaptation and a tool for estimation of interactions between different brain regions during learning.