Objective-To investigate whether upper airway sounds of horses exercising with laryngeal hemiplegia and alar fold paralysis have distinct sound characteristics, compared with unaffected horses. Animals-6 mature horses. Procedure-Upper airway sounds were recorded in horses exercising on a high-speed treadmill at maximum heart rate (HRMAX) under 3 treatment conditions (ie, normal upper airway function [control condition], and after induction of left laryngeal hemiplegia or bilateral alar fold paralysis) in a randomized crossover design. Fundamental frequency, spectrograms using Gabor transform, and intensity characteristics of acquired sounds (peak sound level [soundpeak] and highest frequency of at least –25 dB sound intensity [F25max]) were evaluated. Results-Evaluation of the fundamental frequency of the time domain signal was not useful. Sensitivity and specificity (83 and 75%, respectively) of spectrograms were greatest at maximal exercise, but the exact abnormal condition was identified in evaluation of only 12 of 18 spectrograms. Increased accuracy was obtained using soundpeak and F25max as discriminating variables. The use of soundpeak discriminated between control and laryngeal hemiplegia conditions and F25max between laryngeal hemiplegia and alar fold paralysis conditions. This increased the specificity of sound analysis to 92% (sensitivity 83%) and accurately classified the abnormal state in 92% of affected horses. Conclusions and Clinical Relevance-Sound analysis might be a useful adjunct to the diagnosis and evaluation of treatment of horses with upper airway obstruction, but would appear to require close attention to exercise intensity. Multiple measurements of recorded sounds might be needed to obtain sufficient accuracy for clinical use.