A new method combining both stratification on pre-treatment covariates, and rerandomization when initial randomization does not achieve covariate balance, outperforms existing methods in the context of covariate imbalance.