Efficacy and tissue effects of injection therapy for nonvariceal upper gastrointestinal bleeding were studied in 16 mongrel dogs. The results were compared with those obtained by neodymium-yttrium-aluminum-garnet laser, bipolar electrocoagulation, and heater probe. Epinephrine (1:10,000), absolute ethanol, and 1% polidocanol were used as injection solutions. In acute, severely bleeding experimental ulcers as well as in transected submucosal arteries, injection methods were not as effective as thermal methods in achieving complete hemostasis, although injection therapy, especially with large volumes of epinephrine (1:10,000), very quickly decreased the rate of bleeding. Chronic experiments showed that epinephrine (1:10,000) caused almost no tissue injury, but also did not induce vessel thrombosis. In contrast, absolute ethanol and 1% polidocanol caused tissue necrosis, ulceration, and vessel thrombosis, the former by acute dehydration and fixation of the tissue, the latter by acute edema and subsequent inflammation and sclerosis. These data suggest that although epinephrine injection may slow or temporarily stop bleeding, this modality is not as efficacious as injection with 1% polidocanol or absolute ethanol in inducing definitive vessel thrombosis. In the present experimental conditions sclerotherapy was not as effective as thermal methods in achieving hemostasis. These data also show that injection therapy with 1% polidocanol and absolute ethanol is by no means safer than thermal methods.