A completely structured interview is where the researcher reads out from a set of questions exactly as they are set out and never asks anything else. In other words, it's a questionnaire with an interviewer. So of course very few interviews are completely structured. At the minimum a little prompting for alternatives is allowed.
In a completely unstructured interview what you have is a conversation which can go anywhere. It is hoped this way that a natural conversation leads to avoid influencing the responses beyond how they would be influenced in any other normal conversation. But of course it may be necessary to direct the conversation back to the issues you are concerned with.
So a structured interview allows for some spontaneity and an unstructured interview allows for some redirection. A semi-structured interview has a set of questions or issues to cover but allows for conversation and redirection.
To do an interview needs preparation. First you have to get the confidence of the interviewee. You want their co-operation including a willingness to give full and frank answers to questions. A method of doing this is to begin with light conversation that develops a rapport with the interviewee but you also have to explain what it is all about. To do this you need to be honest about why you are asking them, who else you are asking, what the research is, when and what will be the presentation, what can they see of it, how you will respect their anonymity if this is necessary or otherwise assure them of fairness and accuracy if they are not anonymous.
There are two ways to ask questions, and you should state what the method is to the interviewee. The first method is non-directive. You refrain from making judgements and criticising. You make yourself neutral or slightly sympathetic. You are always polite and helpful. The second way is to get stuck in and be directive or even aggressive. Prepare the ground for this, however, and say that these questions need to be asked, and that you hope to engage in debate. You explain that you are taking on a position, if only for the method of the research. They may like this - think it is a game and see if they can win the argument against you. as part of this approach you can also play dumb, that is from time to time suddenly come across as thick. They're coming at you with an argument, see a weakness and go for it. Good, because you've got better data that way. But you then need to quickly come back or move the conversation on.
If there are several interviewers to one interviewee, this can put the interviewee on the spot. It's like a job interview. It is on the other hand possible for one or more interviewers to carry out a group interview where the interviewees offer each other support and comments.
An interview is necessarily a half way house. It's not as structured or as reliable as a questionnaire, and it is not as data rich and therefore valid as participant observation. But it does allow for something near the reach of a questionnaire in terms of numbers while getting something of the additional data that you do get with participant observation. They are therefore practical and flexible, but there is a sense in which an interview method should not be used alone. Somehow the compromise questions both the reliability of the data - the neutrality and potential repeatability of the data - and the validity of the data in terms of lack of natural development and depth. Try and use interviews along with another method at least.
And in any case people can lie in interviews, they can lead the researcher on, they can have poor memory which they fill in front of the interviewer, and they recall the past in the light of their present day perceptions. We know that people recall the past as rosier than it was. Even British Rail has joined the golden age of the railways now. Also what people say they do is not necessarily what they do. A study of drivers' behaviour that relied on interviews should be laughed out of court! At the same time the interviewer presents his or her cultural baggage to everyone, and questions can be persistently misunderstood. You have to relate to cultures and groups - this is why participant observation is often preferred because at least a culture and group speaks of itself through its normal actions and behaviour, not through answers to questions which may barely be understood.
Although a more or less unstructured interview is set up to allow a natural conversation to occur and thus reveal data in the respondent's own way, the very opposite can happen. It allows too much influence from the interviewer on the giving of the interviewees answers. This is why a pre-determined and up front argumentative approach can be better.
And what if people do answer questions in a way very positive for data collection? The researcher than goes away and classifies and re-represents that data. This process can distort the meaning that the interviewee gave.
Haralambos, M., Holborn, M. (1991), Sociology: Themes and Perspectives, London: Collins Educational, 734-740