In this paper, we tried evolutionary designs of f-θ lenses for a laser printer or a digital copier with a real-valued GA (NDX: Normal Distributed Crossover) and a genotype GA with high mutation. The f-θ lens represented by a B-Spline curve refract a laser beam reflected from rotating mirror to scan a photo sensitive drum with a constant velocity. Fitness is evaluated by accumulated errors between destinations of ray traces and given targets. We compared the performances of two evolutionary methods. Empirical results show that the real-valued GA has converged faster with better accuracy than the genotype GA for the f-θ lens design problem with strong epistasis.