log linearization in dynare

[Hoyd]

So far as I know if a log-linearized equation system is required, then one way is to log-linearize it manually, then put in dynare and begin the model block with "model(liner)", in that way dynare will take the 1st order taylor expension for the linearized equations, will those differ from the original log-linear ones ?

The 2nd way is to express the variables as exp(x) in the original non-linear system and let dynare to log-linearize it, somehow these two ways yield quite different results in terms of the shape of impulse response ... Or there might be something wrong with my understanding?

Thanks for any help provided.