This paper introduces a novel method for automatic age estimation, named AGES (Aging Pattern Subspace), which addresses the unique challenges of aging variation in facial recognition. Unlike other facial variations such as identity, expression, and gender, aging presents three distinctive characteristics: uncontrollable progression, personalized aging patterns, and temporal data. These characteristics make age estimation a complex task. The AGES method models aging patterns, defined as sequences of face images sorted by time, by constructing a representative subspace. The proper aging pattern for an unseen face image is determined by projecting it onto the subspace that minimizes reconstruction error, and the position of the face in that pattern indicates its age. The paper compares AGES with existing age estimation methods (WAS and AAS) and classification methods (kNN, BP, C4.5, and SVM), finding that AGES performs significantly better than all other algorithms and is comparable to human observers. The experiments use the FG-NET Aging Database and the MORPH database, and the results show that AGES effectively handles incomplete data and generalizes well to new subjects. Additionally, AGES can simulate aging effects, which has applications in face recognition across ages. Future work will focus on improving the preprocessing method to consider face size and shape, and exploring the use of AGES in other computer vision tasks, such as pose and illumination variations.This paper introduces a novel method for automatic age estimation, named AGES (Aging Pattern Subspace), which addresses the unique challenges of aging variation in facial recognition. Unlike other facial variations such as identity, expression, and gender, aging presents three distinctive characteristics: uncontrollable progression, personalized aging patterns, and temporal data. These characteristics make age estimation a complex task. The AGES method models aging patterns, defined as sequences of face images sorted by time, by constructing a representative subspace. The proper aging pattern for an unseen face image is determined by projecting it onto the subspace that minimizes reconstruction error, and the position of the face in that pattern indicates its age. The paper compares AGES with existing age estimation methods (WAS and AAS) and classification methods (kNN, BP, C4.5, and SVM), finding that AGES performs significantly better than all other algorithms and is comparable to human observers. The experiments use the FG-NET Aging Database and the MORPH database, and the results show that AGES effectively handles incomplete data and generalizes well to new subjects. Additionally, AGES can simulate aging effects, which has applications in face recognition across ages. Future work will focus on improving the preprocessing method to consider face size and shape, and exploring the use of AGES in other computer vision tasks, such as pose and illumination variations.