Protein complexes play essential roles in many cellular processes, and the arrangement of their subunits is often critical to their function. AlphaFold Database (AFDB) multimer predictions provide valuable structural insights into these assemblies, but interpreting their confidence scores requires care.This webinar introduces the confidence metrics used to assess AFDB multimer predictions. It begins with the foundational scores  - predicted Local Distance Difference Test (pLDDT) and Predicted Aligned Error (PAE) -  before examining five metrics used to evaluate predicted protein complexes: average pLDDT, interface predicted Template Modelling score (ipTM), interaction prediction Score from Aligned Errors(ipSAE), predicted DockQ version 2 (pDockQ2), and Local Interaction Score (LIS).The webinar explains how these metrics are calculated, what they reveal about predicted protein–protein interfaces, and their limitations. In particular, it explores how disordered regions can affect ipTM and how metrics such as ipSAE and LIS reduce this distortion by focusing on confidently predicted interactions.