The grammar of Lojban is based on predicate logic. The majority of the grammar is borrowed from the prior "logical language" Loglan, and some of its features come from Láadan. The characteristic regularity, unambiguity, and versatility of Lojban grammar owes much to modern linguistics and computer programming—resources that were unavailable to the designers of earlier languages. Lojbanist Bob LeChevalier summarized one advantage of Lojban grammar as follows: "Lojban moves beyond the restrictions of European grammar. It overtly incorporates linguistic universals, building in what is needed to support the expressivity of the whole variety of natural languages, including non-European ones."

Lojban texts can be parsed just as texts in programming languages are by using formal grammars such as PEG, YACC, Backus–Naur form. There are several parsers available.

Lojban has 6 vowels and 21 consonants. The phonemes are commensurate with graphemes, which means Lojban has 27 letters (lerfu) each corresponding to a unique phoneme. Lojbanic graphemes can vary in mode; this article employs the Latin alphabet version, which is currently in the most common usage (see Orthography for more detail). The phonemes, on the other hand, are defined solely according to the International Phonetic Alphabet.

The tables below show typical realizations of sounds and the Latin alphabets in Lojban. In all cases except the rhotic consonant the first phoneme represents the preferred pronunciation, while the rest are the permitted variants intended to cover dissimilitude in pronunciation by speakers of different linguistic backgrounds.

Note that simple vowel graphemes are never diphthongized like in English. Diphthongs are always written with two vowel graphemes.

Lojban has 16 diphthongs, vowels that change quality during their emission but always being single syllable nuclei like pure vowels. Unlike English and similarly to languages such as Spanish, diphthongs are not distinct phonemes by themselves but are analyzed as a combination of "semi-vowel + vowel" (or the inverse order). The combinations <ai>, <au>, <ei> and <oi>, for instance, are all realized as the corresponding falling diphthongs. Triphthongs exist as combinations of a rising and a falling diphthong, e.g. <iau>.^{[citation needed]}

The vowels can be either rounded or unrounded. The consonants can be either aspirated or unaspirated, although the voiceless stops /p/, /t/ and /k/ are usually aspirated to some degree. In general, consonants are not palatalized. The affricates /d͡ʒ/, /t͡ʃ/, /d͡z/, and /t͡s/ occur in Lojban (like in English jar, chair, fads, and cats.), but they are not distinct phonemes, contrary to English /d͡ʒ/ and /t͡ʃ/. Each is considered to be a combination of an appropriate stop and fricative phoneme: the sequences <dj>, <tc>, <dz> and <ts>, respectively. The rhotic sounds are all equally acceptable as an identical phoneme. ⟨l⟩, ⟨m⟩, ⟨n⟩, and ⟨r⟩ may be syllabic.

For those who, given their native language background, may have trouble pronouncing (certain) consonant clusters, there is the option of inserting buffer vowels between them, as long as they differ sufficiently from the phonological vowels and are pronounced as short as possible. This mechanism can be described as a form of anaptyxis (vocalic epenthesis). Possible choices include [ɪ] , [ɨ], [ʊ] and [ʏ] (but not [y], which is the rounded counterpart of [i] and thus a valid realization of /i/). The resulting added syllables are completely ignored by the grammar, including for the purposes of stress determination.