The advent of remote work and video conferencing has introduced unique challenges, with “Video Conferencing Fatigue” (VCF) emerging as a prevalent issue. Rooted in increased mental workload, VCF stems from factors like constrained movement, limited non-verbal cues, prolonged gaze fixation, and degraded audiovisual quality. While self-reported measures shed light on user experiences, they often lack real-time applicability. This study investigates the potential of biosignals as objective markers for quantifying mental workload and fatigue during video conferences. An existing dataset of 127 participants was analyzed to examine physiological dimensions-including heart rate variability (HRV), electrodermal activity (EDA), eye movements, and facial expressions—captured during systematically manipulated audiovisual quality levels across different conversational tasks. Results reveal compelling evidence to support three hypotheses: (1) Instead of simply increasing over time, perceived fatigue is significantly influenced by task complexity; (2) lower audiovisual quality substantially elevates mental workload; and (3) physiological signals like HRV decrease, EDA nonspecific responses increase, and eye metrics such as saccade rates and pupil dilation correlate with VCF. We further present a multimodal machine learning ensemble based on these biosignals for automatically classifying fatigue. The model achieved an F1 score of 0.6 and an accuracy of 0.71 in distinguishing fatigued from non-fatigued states, with eye-tracking features emerging as the most predictive. These findings highlight the impact of task complexity and audiovisual quality on VCF and demonstrate the potential of psychophysiological measures for real-time fatigue monitoring and adaptive video conferencing to mitigate its effects.