If you are looking for a phonetic basis for thinking that you have [gw] versus [gʷ], you can listen for an effect on the preceding segment, where the end of the previous vowel is more likely to show signs of partial rounding before [gʷ], compared to if you have a sequence [gw]. I said "likely", so you can't use this as an absolute distinguishing feature.
Generally, you have to rely on phonological tests. One is sonority-sequencing in the syllable – the tendency for obstruents and sonorants to be ordered so that the sonorant is nearest to the syllable peak (vowel). A word like [bəkw], qua consonant-glide sequence, is less likely than a single consonant at the end of the syllable i.e. [bəkʷ]. Again, this is a tendency, and one would be hard-pressed to insist that syllable-final consonant plus glide sequences are universally impossible.
Another argument could be pattern of reduplication. If a language has consonant plus glide sequences = rounded consonants, and also consonant plus liquid, plus also has a pattern of CV reduplication, then the application of reduplication to supposed Cw could be a good diagnostic. Thus if /plam/ → [pa-plam] and /kwat/ → [kakwat], you have evidence for a two-segment analysis as velar plus glide, but [kʷakʷat] would be evidence for a single round segment. Again, this is a (strong) "tend" argument (there is an alternative analysis invoking contrastive structure to the onset, if you believe in such things).
In Berber and Ethiopian Semitic languages (and maybe Moroccan Arabic, details are not clear to me), the "kw" sequences are more clearly secondary articulations on consonants, based on phonological pattern.