Abstract: Based on the historical wet/dry grade series of North China, back to 1261 AD, some nonlinear features of the regional wet/dry fluctuations are analyzed. The spectral structure of the wet/dry system is quite similar to that of the chaotic model such as the Lorenz's. A special error-controlled method is designed to calculate the attractor's dimension, which is then estimated as a fractal one larger than 4. The smoothed series that mainly represents the 102-year-scale fluctuations tends to have a smaller dimension about 3. 5. The local Kolmogorov entropy is analyzed and a 4-year-predictable time is estimated for the wet/dry fluctuations. The K-entropy of the smoothed series is small, but that for those ‘drier' phasepoints seems much smaller. This implies that at the long time-scale, the climate prediction in the drier period is more difficult than that in the wetter one.