Canine influenza virus (CIV) H3N8 was first detected in racing greyhounds in Florida in 2004 and subsequently spread to shelter populations in New York and Florida. The disease is supposedly spread by direct contact, deposition of infectious droplets on the nasal or oral mucosa, or inhalation of infectious particles. Serum and nasal swabs were collected at intake and discharge from 5,160 dogs in 6 shelters in the U.S. (CA, CO, FL, NY, SC, TX) over a 2-year period. Nasal swab positivity varied among the geographic regions (NY, 4.4%; CO, 4.7%; SC, 3.2%; FL, 1.2%; CA and TX, 0%). When data were compared at intake and discharge, trends varied among shelters. For example, in Colorado, 6 times more dogs were shedding virus at discharge (8.9%) than at intake (1.4%). In contrast, in New York, dogs were more likely to be shedding at intake (6.9%) than at discharge (1.1%). Data revealed that the disease demonstrated an ebb-and-flow pattern where outbreaks occurred in 1 shelter and 1 region and waned before the disease emerged in another shelter in another region. Dog-to-dog contact and number of days in the shelter were contributory factors for CIV seropositivity and shedding. Data indicated that community dogs were a source of CIV introduction into a shelter and that once established, dog-to-dog transmission maintained the virus within the shelter.