Objective Effective connectivity differences between control and dyslexic children are explored by a modeling approach. Dynamic Bayesian networks (DBN) are used to model dynamic, nonlinear, non-stationary and probabilistic nature of the human brain connectivity. Methods EEG data from 23 learning disability and 23 healthy children during hard and easy word reading experiments is used. EEG’s high temporal resolution is combined with the nonlinear, non-deterministic and dynamic behavior of discrete DBN. Quantization is done appropriately so that data loss is minimized. Stationary windows of the time series are considered to handle non-stationary behavior. Modeling is done separately for control and dyslectic groups, for resting and reading states, for easy and hard word reading experiments. The resulting models are analyzed statistically to provide the similarities and differences between control and dyslectic groups. Results Significant differences of connectivity are observed between resting and reading states for both classes. The behavior of connectivity changes from resting state to reading state are similar for easy and hard word reading tasks but different for the control and dyslectic classes. Conclusions Significant differences are observed between the two classes of subjects. Key message DBN is very important for brain connectivity modeling, when appropriate modifications are included.