R help - Feb 2016 - lm-step: use train-data model to validate test-data

I have model-data named as: model that is split as model.T(train) and
model.V(test or validation). The least square model (from lm to step) is built
withmodel.T and I like to see how model.T is robust by comparing predicted
model.V toactual model.V. How do I get score for model.V based on model built on
model.T? The code highlighted below does not get what I expected.Please advise!
Thanks!
? # score the model 
? score.T <- data.frame(predict(step, model.T))? # get predicted score for
train data
? score.V <- data.frame(predict(step, model.V))? # for test data but seems
incorrect
? # get the actual values 
? actual.T <- data.frame(model.T$sales) 
? actual.V <- data.frame(model.V$sales) ? # comparison for model.T
? comp.T=cbind(actual.T,round(score.T,digit=2))
? plot(comp.T)? # comparison for model.V (use Model.T to predict Model.V for
true validation
? comp.V=cbind(actual.V,round(score.V,digit=2))
? plot(comp.V)
	[[alternative HTML version deleted]]