Variation is a characteristic of language: there is more than one way of saying the same thing in a given language. Variation can exist in domains such as pronunciation (e.g., more than one way of pronouncing the same phoneme or the same word), lexicon (e.g., multiple words with the same meaning), grammar (e.g., different syntactic constructions expressing the same grammatical function), and other features. Different communities or individuals speaking the same language may differ from each other in their choices of which of the available linguistic features to use, and how often (inter-speaker variation), and the same speaker may make different choices on different occasions (intra-speaker variation).

While diversity of variation exists, there are also some general boundaries on variation. For instance, speakers across distinct dialects of a language tend to preserve the same word order or fit new sounds into the language's established inventory of phonemes (the study of such restrictions known as phonotactics, morphotactics, etc.); however, exceptions to these restrictions are possible too. Linguistic variation does not equate to ungrammatical usage of the language, but speakers are still (often unconsciously) sensitive to what is and is not possible in their native lect.

Language variation is a core concept in sociolinguistics. Sociolinguists investigate how linguistic variation can be influenced by differences in the social characteristics and circumstances of the speakers using the language, but also investigate whether elements of the surrounding linguistic context promote or inhibit the usage of certain structures.

Variation is an essential component of language change. This is because language change is gradual; a language does not shift from one state to another instantaneously, but old and new linguistic features coexist for a period of time in variation with each other, as new variants gradually increase in frequency and old variants decline. Variationists therefore study language change by observing variation while a change is in progress. However, not all variation is involved in change; it is possible for competing ways of "saying the same thing" to coexist with each other in "stable variation" for an extended period of time.

Studies of language variation and its correlation with sociological categories, such as William Labov's 1963 paper "The social motivation of a sound change," led to the foundation of sociolinguistics as a subfield of linguistics. Although contemporary sociolinguistics includes other topics, language variation and change remains an important issue at the heart of the field.

Studies in the field of sociolinguistics typically take a sample population and interview them, assessing the realisation of certain sociolinguistic variables. Labov specifies the ideal sociolinguistic variable to

Phonetic variables tend to meet these criteria and are often used, as are morphosyntactic variables, morphophonological variables, and, more rarely, lexical variables. Examples for phonetic variables are: the frequency of the glottal stop, the height or backness of a vowel or the realisation of word-endings. An example of a morphosyntactic variable is the frequency of negative concord (known colloquially as a double negative). Two well-known and frequently studied morphophonological variables are T/D deletion, the optional deletion of the sound /t/ or /d/ at the end of a word, as in "I kep' walking" (Wolfram 1969; Labov et al. 1968); and the ING variable, the optional pronunciation of -ing at the end of a word as -in', as in "I kept walkin'" (e.g. Fisher 1958; Labov 1966/1982; Trudgill 1974).

Analyzing sociolinguistic variation often involves the use of statistical programs to handle its multi-variable nature. One essential part of the methodology is to count up the number of tokens of a particular variant and compare it to the number of times the variant could have occurred. This is called the "Principle of Accountability" in Tagliamonte (2012). Comparing the tokens to the total number of words in a corpus or comparing one corpus to another leads to erroneous results. This count of the possible occurrences can be difficult at times because some variants alternate with zero (such as relative pronouns that, who, and zero).