The ability to sense and respond to temperature is universally important for the wellbeing of motile animals. Navigation in environments with variable temperature (thermotaxis) is built from signal generation in sensory neurons, signal processing in the brain, and signal conversion to physical muscle action. Working in the Drosophila larva model system, we explore the response to temperature at the scales of populations, individual animals, neuronal circuitry, and molecules. We find that response to heating is independent of response to cooling at all these levels, as the two distinct sensorimotor transformations operate in parallel. Modeling and Monte Carlo simulations using empirically-derived rules for behavioral mode transitions demonstrate that realistic movement can be recovered with a small number of parameters, and show the benefit of randomness in navigation decisions. Behavioral tracking of individual larvae establishes navigation patterns and shows that thermotaxis is specifically an aversive rather than attractive response to both heating and cooling. We employ an improved device for generating rapid changes in temperature to subject larvae to random noise stimuli, then generate mathematical filters that can probabilistically predict the larvae’s responses at high and low temperatures. At the cellular level, we measure in vivo the activity of separate cool-sensing and heat-sensing neurons and identify downstream connectivity that points towards a more complete understanding of complex brain circuits. Finally, by using the above behavioral and neurophysiological methods, we uncovered the mechanism of cooling sensation, showing that two ionotropic receptors (IRs) are required for navigation and detection in adult flies and larvae, and also identifying candidates for heat sensation in the larva. The quantitative approaches used here should also be applicable to work in other sensory modalities and other model systems seeking to characterize sensory and circuit-level steps that lead to behavior.