This study aims to develop and validate an age-appropriate, easily manipulated, and robot-based assessment named Computational Thinking Observational Scale (CTOS) for 3- to 5-year-olds. The study sample included 46 preschoolers (M[subscript age] = 3.57 years, SD = 0.36). Expert judgment confirmed excellent content validity, yielding a Scale-level Content Validity Index of 1.00. Criterion validity showed a strong positive correlation between CTOS scores and interaction quality in children's coding with robots (r = 0.81), with significant correlations across all dimensions, particularly "debugging" (r = 0.86). Inter-rater reliability, assessed using the Intraclass Correlation Coefficient, was 0.89, demonstrating good reliability. Internal consistency was high, with a Cronbach's alpha of 0.94 for the full scale and split-half reliability of 0.89. Gender differences revealed that boys scored higher on "debugging" and "expressing & creating" dimensions, while age analysis indicated significant score increases with age, especially in "hardware" and "choices of conduct" dimensions. Socioeconomic status showed no significant correlation with CTOS scores. These findings support CTOS as a reliable and valid tool for evaluating children's CT in coding robotic interactions. Implications and limitations of CTOS are also discussed to shed light on the future directions of CT assessment and CT education in early childhood.