Procedures for predicting dementia levels were developed using features of waveform shapes of the pupillary light reflex (PLR) in response to chromatic light pulse stimuli. Features of waveform shapes of PLRs were extracted using a functional data analysis technique, and some irregular signal responses used to make the predictions were detected. The coefficient vectors of B-spline basis functions were introduced, and the number of basis functions was controlled to produce an optimised model. The prediction procedure for dementia levels was developed using the features extracted from PLRs during two types of observations. Results showed that some differential waveform shapes of PLRs are based on physiological characteristics contributed to the prediction.