Abstract Alzheimer’s disease and related dementias, a leading cause of disability among older adults, has become a critical public health concern. The clock-drawing test (CDT), which asks subjects to draw a clock, typically with hands showing 11:10, has been widely used as a screening tool to detect dementia in clinical research and surveys. The clock-drawings are often coded into binary (e.g., normal vs. abnormal) or ordinal scores. A limitation in large-scale studies is that the manual-coding of CDT could result in biases if coders interpret and implement coding rules differently. Several small-scale studies have explored the use of machine learning methods to automate CDT coding. Such studies either have had limited success with ordinal coding or have used methods that are not designed specifically for complex images. This study aims to create and evaluate an intelligent CDT-scoring system that automatically codes ordinal CDT scores using deep learning neural networks (DLNN) methods including resnet101, EfficientNet-B3, and Visual Transformer. We used a large repository of CDT images from the 2011-2019 National Health and Aging Trends Study, a panel study of Medicare beneficiaries ages 65 and older. Results show that 1) DLNN has achieved a high scoring accuracy; 2) DLNN generates more consistent and accurate scores than trained lay coders, compared to expert coders with a neuropsychology background; 3) DLNN-coded CDT scores are highly correlated with self-reported dementia diagnosis and performance-based assessments of memory. This study offers a model for automating coding of other drawing tests used to evaluate a variety of cognitive functions.