Hi Raj,

I have never use RandomForest algorithm (I work in potential distribution modeling with MaxEnt and variable importance for the variable importance, but using different statistical aproach GNESFA/MADIFA/ENFA). For what purposes you want to use randomForest, niche modeling?, neurosciences?... Take in account that I am working with nniche modeling and ecology and not with other branch of sciences, so maybe my position is deeply biased.

This is a link about the process of variable importance in RandomForest that sound very similar to other approach, but in this approach lies in comparison of Z-score and takes in account interactions among the variables. This estimates seems to be sensitive to the sample number (n) and the number of trees in the regression-classification process.

http://www.statistik.uni-dortmund.de/useR-2008/slides/Strobl+Zeileis.pdf

I hope this link provides you some helpful information,

Luís Enrique.

The main purpose of Random Forest Algorithm is to give  variable importance/relevance in a dataset, it's perform usually before a clustering or classification. It permits to determine weights of features before clustering/classification.

in R software and "caret" package you can use varImp() function