In order to collect sufficiently many samples and keep their distinguishability in online sketchy symbol recognition, this paper proposes a detector-generation based clonal selection algorithm and an evaluation method. The algorithm generates detectors with an r-contiguous-bits unchanged rule (r-CBUR) and a p-receptor editing to search in a wide feature space and try to avoid local convergences. Hand-Written Chinese characters are selected as experimental samples, for which the influence of the training parameters is analyzed. The experimental results show the improvements of the training process and the classification results of sketchy symbol recognition.