Disorder prediction for short peptides is important and difficult. All modern predictors have to be optimized on a preselected dataset prior to prediction. In the succeeding prediction process, the predictor works on a query sequence or its short segment. For implementing the prediction smoothly and obtaining sound prediction results, a specific length of the sequence or segment is usually required. The need of the preselected dataset in the optimization process and the length limitation in the prediction process restrict predictors' performance. To minimize the influence of these limitations, we developed a method for the prediction of intrinsic disorder in short peptides based on large dataset sampling and statistics. As evident from the data analysis, this method provides more reliable prediction of the intrinsic disorder status of short peptides.