Animals can move towards or away from an odorant. Such chemotaxis has been used as a paradigm for learning when coupled with pre-exposure to the sensed odorant. Here we develop an assay for the nematode C. elegans that avoids the typical use of chemical or physical immobilization when measuring the response of worms to odorants. Using two sets of rectangular arenas that are oriented such that worms in one set must move in the opposite direction to worms in the other set for the same response, we found that unfed worms show reproducible movement towards the odorants butanone and benzaldehyde, and away from the odorant nonanone. In addition to the use of opposing orientations to control for gradients of unknown cues outside the arena, we introduce a measure of dispersal to control for locomotion defects and unknown cues within the arena. Since this assay avoids the use of paralytics or physical constraints, it is useful for the analysis of graded responses to a variety of chemicals and the discovery of underlying molecular mechanisms. Using this setup, we found that pre-exposure of unfed worms to butanone to induce an association of starvation with butanone resulted in different extents of such associative learning during different trials - from no learning to learned avoidance. Given this variation in associative learning despite the artificially controlled lab setting, we speculate that in dynamic natural environments such learning might be rare and highlight the challenge in discovering evolutionarily selected mechanisms that could underlie learning in the wild.