Representations are ubiquitous in chemistry. They are part of the chemistry language instructors use to communicate chemistry phenomenon to students. Literature calls in support of learning with multiple representations, but there is a pre-requisite for students to learn from a single representation. In this exploratory study, 1086 students in second semester general chemistry were randomly assigned to one of four representations showing bonding of sulfur dioxide: chemical formula, Lewis dot structure, an image of a ball and stick model, or an image of a space filling model. Students were asked to predict chemical properties of sulfur dioxide: relative bond length, molecular polarity, and the strongest intermolecular force with a water molecule. Using the lens of Multimedia Learning Theory on Learning with Text and Visual Representations, analyses of students' prediction of chemical properties and the features cited when making predictions was conducted. Effect sizes were used to describe variations among representations in terms of how students predicted bond length, polarity and intermolecular forces. Meaningful differences were found across representations in students' ability to correctly predict relative bond length and molecular polarity. These explorations generated the following hypotheses: (1) chemical formula leads students to depend on chemical conventions, (2) Lewis dot structure hinders predicting polarity when it shows an inaccurate shape, and (3) visual representations of ball and stick and space filling cue students to rely on visual estimations more than the other representations. Upon further testing, these hypotheses can inform instructors how to introduce representations and in the decision-making process of which representations to use to convey or assess a specific chemical property.