Background and Context: Current programming learning research often compares novices and experienced programmers, leaving early learning stages and emotional and cognitive states under-explored. Objective: Our study investigates relationships between cognitive and emotional states and learning performance in early stage programming learners with varying expertise (12 novices and 11 beginners; N = 23). Method: Measures such as electrodermal activity, automated facial emotion recognition, and pupil diameter were collected to study cognitive and emotional states during a programming task. We examined how these states correlated with performance and the influence of expertise on these correlations. Findings: Multiple regression analysis on panel data (n = 690 observations) suggested fear negatively impacts programming performance, irrespective of expertise level. Meanwhile, programming expertise positively moderated cognitive effort applied to the task. Implications: Our findings highlight the importance for educators to design early programming experiences that anticipate and reduce fear and cognitive overload, promoting better learning outcomes.