<p>Recognizing masked faces is a challenge. Researchers have explored congruency-based approaches to improve face matching, with promising results. Here, we investigated whether congruency between the encoding and the retrieval conditions can improve masked face recognition when only the eyes are visible under conditions of high and low memory load. Additionally, we explored whether the advantage of congruency varied as a function of general face recognition ability. In three experiments (total <i>N</i> = 316), participants completed a face recognition task that manipulated the congruency between encoding and retrieval conditions. In congruent sets, the images featured either a full face or a partial face at encoding, paired with a full or partial face, respectively, at retrieval. In incongruent sets, the images paired a full face at encoding with a partial face at retrieval or a partial face at encoding with a full face at retrieval. The Cambridge Face Memory Test served as a measure of general face recognition ability. The results supported the hypothesis that contextual congruency improves face recognition (<i>η</i>_<i>p</i>^<i>2</i><i> ≥</i> .46), and the effect remained consistent across high- and low-ability face recognition performers. Additionally, memory load shaped the confidence-accuracy relationship, such that confidence was a reliable predictor of accuracy under conditions of low (but not high) memory load. These results show that focusing on originally encoded facial features significantly improves masked target recognition, aiding law enforcement in identifying masked perpetrators and enhancing public safety.</p>