One of the most challenging issues faced by many elders is the over-decreasing independence mainly caused by impaired physical, cognitive, and/or sensory abilities. Activity recognition can be used to help elders live longer in their own homes independently, by providing assurance of safety, instructing performance of activity and assessing cognitive status. In this work, we propose to discover both intra- and inter-activity association patterns from daily routines of elderly people. Specifically, a data mining method is proposed to extract the most frequent sequential sequences of steps inside each individual activity (i.e., intra-activity pattern) and activities (i.e., inter-activity pattern) of a set of daily activities. These patterns can then be used to model human daily activities for activity recognition purpose, or to directly instruct/prompt elders with impaired memory when they perform daily routines. The experimental results conducted on two individuals' datasets of daily activities show that our proposed approach is workable to discover these association patterns