Think about it this way.
If Oracle happens to think (based on the stats gathered etc) that your query will return "most" of the rows from the table (how much exactly is "most" is again a decision made by the optimizer), it will resort to a full table scan because its faster to just get all the rows and reject the ones that are not required than to go through the index and find out the ones that are required. That way it can save the additional step of accessing the index which in itself is a data structure that needs to be traversed.
Regards,
Arijit