(Japanese on'yomi 音読み, literally "sound-reading"). The "Chinese" pronunciation of a Japanese kanji.
Japanese lacked a writing system until it came into contact with Chinese writing in the fourth century. In Chinese, every character has a specific meaning and a specific pronunciation (and the pronounciation can often be guessed from components of the character, c.f. radical). When these characters were imported into Japanese, they were used in two distinct ways: to write the Japanese word with the same meaning as the character, and to denote sounds through their pronunciation.
Consequently in modern Japanese, every kanji can be read in two different ways: either as something similar to the Chinese pronunciation, the on-reading, or as something similar to a native Japanese word, the kun-reading. Which reading is used varies from word to word. For example the character 人 is read with its kun-reading "hito" in hitogomi 人込み 'crowd', but with the on-reading "jin" in jinsei 人生 'life'. This state of affairs is an inconvenience to (at least) students of the language, since it means that it often isn't possible to know the pronounciation of an unknown word in a text without looking it up.
There is an additional complication because characters were imported into Japan over a long period of time, from areas speaking different Chinese dialects, and a given character may therefore have more than one on-reading. The on-readings are divided into three "waves":
Japan first came into contact with Chinese writing via the ancient Korean kingdom (Baekje) in the middle of the fourth century. Korea in turn was mostly in contact with the region of China around downstream Yangtze river, and the pronounciation of the characters was taken from the dialect spoken there. The Japanese readings originating in this period are called go-on (呉音). Because Buddhism was introduced to Japan at this time, Buddhist terminology and words derived from it use go-on readings (e.g. ningen 人間 'human').
In 589, China was reunited under the Sui dynasty, which was followed in 618 by the Tang dynasty. The national capital was established in Xi'an, and Japan started sending emmisaries there. The emmisaries had to learn the dialect spoken in the area, and then brought it back to Japan. These readings, coming from Tang-dynasty Xi'an, are called kan-on (漢音). At this time, classical Chinese poetry and literature was held in high regard among the upper classes, and words deriving from such sources use kan-on readings. In modern Japanese, kan-on readings are much more common than go-on, but this is relatively a modern state of affair. It began with the Meiji era, which saw the diffusion of classical Chinese literature to broader layers of society, and also the creation of many new compound words (using kan-on readings) to describe newly introduced Western things.
The above two waves account for most of the on-readings, but there has been a steady trickle of imported Chinese readings up until modern times. Somewhat arbitrarily, all readings imported after the beginning of the Kamakura era (when the first shogunate was established) are refered to as tou-on (唐音). The majority of these are from the period 1200-1600, and correspond to Chinese pronounciation during the Song, Yuan, and Ming dynasties. This had been a period of great language change in China, and tou-on readings are closer to modern Mandarin than go-on or kan-on readings are. Tou-on readings are relatively uncommon, and hence appear strange or irregular. Unlike the previous two waves there was never a systematic effort to associate characters with pronounciations for these readings. Instead they are the result of borrowing both the spelling and pronouciation of individual Chinese words as they were encountered. They can be found for example in the names of items that were first traded to Japan in this period (e.g. futon 布団; isu 椅子 'chair'; udon 饂飩).
Due to the difference in phonetics, differences between Chinese dialects, and langauge changes, the on-readings are very far from modern Chinese. However, someone familiar with the typical sound-changes can often make an educated guess about the on-reading of a character based on its Chinese pronounciation, and they give valuable information to historians of language when reconstructing historical Chinese.
In modern Japanese, a standard way to form compound words it to concatenate a string of kanjis pronounced by their on-readings. The frequency of this type of word varies a lot depending on the kind of text, however. Words that are made up of on-readings are known as kango (漢語) and are a distinctive feature of written formal Japanese. Newspaper articles and law texts will consist almost only of kango, while they are less common in spoken language.
Sources: The information about go-on, kan-on, and tou-on readings were taken from Ichirou Shida's webpage at http://homepage1.nifty.com/forty-sixer/kotoba.htm