This paper aimed to investigate contributing factors in length of stay at various types of tourist attractions based on cellular data. Accelerated failure time (AFT) models with three error terms distribution were compared: Weibull, log-normal, and log-logistic. The model with frailty was further used to accommodate unobserved heterogeneity. Potential factors related to attractions and trip characteristics are selected. A case study on Yi-Lan County, Taiwan, where has a variety of sightseeing spots attracting visitors, was conducted. The estimation results show that the Log-normal AFT model with frailty performs best. The key contributing factors in length of stay are city of origin, travel distance, entrance time, travel mode, and sequence of visited spots.