Dear Amir,

you have to always check the p-value

You can do Likelihood Ratio Tests (reported as "Chi squared") to compare nested models with vs. without a term, or with vs. without a whole level. You can also use information theoretic model fit indices (AIC, BIC) and compare alternative (not necessarily nested) models.

(If using lme4 in R, just use the anova(model1,model2) function to get these.)

To approach a concept like "model significance", it's not necessarily that meaningful but you could compare a "nil" model (outcome ~ a constant) vs. your full hypothetical model. But remember that if your model is significantly better than a nil model, does not mean that every term is significant in it!