http://www.johndcook.com/bias_consistency.html

Unbiased: estimate theta multiple times, then their expected average is the parameter of population.

Consistent: estimator converges to parameter of the population as the sample size got bigger.