Step width is an important indicator of gait stability and fall risk, and continuous assessment in daily life is required. Although step width can be estimated from gait data collected using inertial measurement units (IMUs), the minimal IMU placement suitable for continuous use in daily life has not been clarified. In this study, we evaluated the step width estimation accuracy across different sensor configurations using gait data collected from IMUs attached to the waist and shank, and, for the first time, an IMU embedded in the insole. Gait data from 24 healthy males were used, and step width was estimated using a deep learning model to evaluate the performance of each sensor configuration. Overall, the combination of insole and waist IMUs showed the highest accuracy, achieving a mean absolute error (MAE) of 39.95 ± 13.39 mm, and among single sensor configurations, the insole IMU achieved the best performance with an MAE of 45.41 ± 14.87 mm. Considering ease of attachment and practical use in daily life, the insole IMU may be a promising sensor configuration that enables high accuracy step width estimation with minimal burden.