If your dependent variable is discrete, you should be using glm with a poisson model. You can use lm but you're obviously violating assumptions.
See example: http://www.ats.ucla.edu/stat/r/dae/poissonreg.htm
And I concur with the post by TBSRounder.