Dear Bidhan, I have following issues with DCM analysis in my experiment: Problem 1 : Models for optimal intrinsic effective connectivity: we have defined 4 DCMes to see the best intrinsic effective connectivity between 3 RIOs. From BMS analysis, model 4 turned out to be the best model. Then we have averaged parameter of model 4 across all subjects (BPA) . But for one connection the estimated probability (DCM.Pp.A) is less than 0.80 ( I think this mean probability is 80 %). Question : Does it ( ie 80 %) mean this intrinsic connectivity is not significant and need to avoid for the further analysis (meaning we are trying to see the effect of six different visual pictures on connectivity after finding the optimal model from intrinsic effective connectivity ) even it is a best model from BMS analysis ? If you remove this connection in DCMes it will be another model ( model 3 instead of model 4). The posterior probability of .8 means - as you say - there is an 80% confidence that this connection exceeds its prior expectation. While this is less than 95%, the fact that the full model (with this connection) had more evidence than the reduced model (without this connection) means that you can be fairly certain this connection is evidenced by the data. I would therefore use model 4. Problem 2 : To see the modulatory effect of all six different picture types ( we have six different kinds visual pictures) we have redefine model 4 with modulation (matrix B) each input types ( as shown in fig) . and BPA was done. But all estimated probability (DCM.Pp) for matrix B are less than 0.75 ( i guess it mean the probability is 75% ). is it still ok to report? If the model with modulatory (picture category) effects has more evidence than a model without these modulatory effects, then you can certainly report the modulatory effects. Your inference is at the level of models - not about any particular modulatory effect on the connections. The simplest thing to do would be to report the log evidence for models with and without specific modulatory (picture category) effects and then discuss the parameter estimates (DCM.Ep.A) quantitatively. Notice that the parameter estimates are not the same as the posterior probabilities (DCM.Pp.A). It may be that having six modulatory parameters means your model is too complex. If you can categorise your pictures - say in terms of two factors (e.g., animate versus inanimate and positive versus negative valence), you could reduce the number of modulatory effects to 2 and these two modulatory effects (B parameters) may be estimated more efficiently - leading to higher posterior probabilities. I hope that this helps - Karl