Thanks to their high strength to mass ratio, composite materials are now widespread in the aerospace industry. Nevertheless, this type of material is sub- ject to internal damages like delamination. In order to detect and localize these damages, robust and precise Structural Health Monitoring algorithms exist for this purpose and have been validated experimentally. However, in order to avoid struc- tures catastrophic failures and to estimate their residual life, there is still a huge need of reliable damage size assessment methods. In this paper, a damage quanti cation method is proposed. This strategy is based on the extraction of a damage size sen- sitive feature computed from damage imaging results. Here damage imaging stands for methods that use ultrasonic Lamb waves-based map of damage localization like- lihood index. This feature is extracted from each labelled example of a training set in order to infer a mathematical model used to predict the area of a delamination of unknown damages. The proposed method is successfully validated on experimental data carried out on CFRP plate samples equipped with a piezoelectric transducers network.