The response surface methodology is a useful method to find the optimum response in an experiment. In this work, a new test statistic with only few replicates at the center point for curvature detection is discussed. The performance of the new statistic is investigated through simulation. In the second part of this work, when the response variable is of the proportional type taking only values between 0 and 1, some analysis methods are compared based on the predicted variances on the design region.