Data which are closest to the ultimate substrate of linguistics are records of natural speech events. Beside these, responses to elicitation, to stimuli and to experimental tasks may be recorded.
Both technical possibilities and quality requirements have increased over the decades. In the first quarter of the 21st century, the best option is to take a video record of the speech event in question, no matter whether it is an entire narrative, a dialog or polylog, a response to a linguistic or non-linguistic stimulus, informal or as part of an experiment. It is worth the effort to mount a good camera with microphone on a tripod and to videotape the entire performance of a speaker or his interaction with others. One might think that, once the concern is with oral language, an audio recording should be sufficient. However, video records are indispensable for many purposes. For instance, an accurate phonetic transcription may depend on it. Moreover, natural texts are full of deictic expression which are accompanied by gestures. If the gestures are not seen, it is sometimes impossible to know what the speaker is referring to. The same is true a fortiori if non-linguistic stimuli or responses are involved in a task.
Whenever the researcher is not a native speaker of the language recorded or if the record is to be made available to non-native speakers, the first thing to do is a transcription of the recorded text and a translation. Failing this, the record will probably be of little use at a later time.