Odor information is represented by both intensity and timing of neuronal activity in the olfactory bulb, but origins and roles of the temporal pattern remain enigmatic. Here we addressed this issues using in-vivo two-photon calcium imaging of the mouse olfactory epithelium and olfactory bulb. We found that nasal airflow produces widespread responses in olfactory sensory neurons, and thereby entrains respiration-locked theta oscillations in mitral/tufted cells. Most glomeruli demonstrated the theta oscillations, but their oscillation phases relative to the sniff cycles were glomerulus-specific. Changes in sniff speed or frequency changed activity intensity, but had minor effects on the relative phase. In contrast, odor stimuli generated odor- and glomerulus-specific phase shifts, indicating that phase information distinguishes odor responses from airflow responses. Notably, during odor sampling across multiple sniffs, the intensity of odor-evoked activity dynamically evolved over time; however, the phase code remained constant across sniffs. Thus, the phase code can more stably represent odor identity than the rate code. The airflow sensation by OSNs is essential for the phase odor coding, because the phasic representation of an odor was impaired under continuous airflow condition. Consistent with this observation, the phasic odor responses were less evident in the olfactory epithelium, suggesting that the oscillatory responses are mainly produced in the olfactory bulb circuits. Together, our results demonstrate that glomerular theta oscillations driven by the nasal airflow are the basis for reproducible perception of odor information. This phase coding may be a sampling mode-invariant and noise-resistant odor coding strategy.