Dear Yu Lei, This is a very open-ended question and you are likely get quite different answers from different researchers. Here's my 2 cents. The field seems to have settled on (i) using linear systems models (GLMs) to quickly scan the whole brain looking for any effects related to the experimental paradigm. This will then produce a small number of candidate regions that one can then (ii) use (computationally expensive) nonlinear models, with hidden variables describing neuronal activity, to model interactions at the neuronal (rather than hemodynamic) level. Best, Will. ________________________________ From: SPM (Statistical Parametric Mapping) [[log in to unmask]] on behalf of Yu Lei [[log in to unmask]] Sent: 29 March 2012 17:28 To: [log in to unmask] Subject: [SPM] Question about linear model and nonlinear model Dear SPM experts, I'm a student in the neuroimaging data analysis. Recently, I've found that in the study of the relationship between stimuli and hemodynamic response, someone use linear system models, such as GLM and FIR model, and the others use nonlinear system models, such as state-space model. I wonder if someone can tell me some advantages and disadvantages about these two different models? Thanks, Yu Lei Zhengzhou Science and Technology College, China