In the design of experiments for response surface analysis, sometimes it is of interest to estimate the difference of responses at two points. If differences at points close together are involved, the design that reliably estimates the slope of response surface is important. This idea was conceptualized by slope rotatability by Hader & Park (1978) and Park (1987). Until now, second order polynomial models were only studied for slope ratatability. In this paper, we will propose the necessary and sufficient conditions for slope rotatability over all directions for the thired order polynomial models in two, three and four independent variables. Also practical slope rotatable designs over all directions for two independent variables are suggested.