Many developers are defining a repository interface with at least a method which returns an IQueryable, especially when they define a generic repository. It's quite tricky to see it, but if you're doing that, you're using the Repository Pattern the wrong way.
First aspect is that having a method returning a IQueryable is a case of leaky abstraction. But you can counter that by saying that IQueryable is part of BCL and Linq is part of C# . Fair enough!
However the mistake is not the IQueryable itself, but its purpose. An IQueryable is pretty much a query builder and it's great for that. A repository method returns some business object (model). Can you spot the difference? You ask the repository to give you a model. How the repository comes up with the model is its problem.
The point is that using IQueryable, you're asking for a query builder and not for a model. A query builder specifies how the query should be, how to get the data. Then why are we using a repository? Isn't the repo's job to know how to get the thing we want? We're using the repository because we want to specify what to get, not how to get it.
If you want to have full control over how to get the model, then you don't need a repository. Use IQueryable or some other builder for the purpose. But if you don't care how or where from the model is retrieved, use a repository whose methods return only a model or an enumerable of models. Let the repository do its job of hiding the implementation details. Use it as an abstraction: tell it the what, not the how.