Degradation tests are widely used to assess the lifetime distribution of highly reliable products that are not likely to fail under a traditional life test. To conduct a degradation test, three key variables, observed frequency, sample size and termination time shall be given a special consideration. Tseng & Liao (1998) proposed a method to design a degradation test. Under the constraint that the total experimental cost does not exceed a pre-determined budget, the optimal decision variables are obtained by minimizing the variance of □ (the estimated pth percentile) of the product’s lifetime distribution. The quasi-linear degradation path proposed by Tseng & Liao (1998); however, may lead Var(□) to be extremely large. To overcome this difficulty, motivated from a real light emitting diode (LED) data, we propωe a nonlinear diffusion process to model the degradation path and the corresponding optimal design can be solved successfully. Finally, we proposed use an example to illustrate the difference between these two approaches.