I assume that with interface you mean a C++ class with only pure virtual methods (i.e. without any code), instead with abstract class you mean a C++ class with virtual methods that can be overridden, and some code, but at least one pure virtual method that makes the class not instantiable. e.g.:

class MyInterface
{
public:
  // Empty virtual destructor for proper cleanup
  virtual ~MyInterface() {}

  virtual void Method1() = 0;
  virtual void Method2() = 0;
};


class MyAbstractClass
{
public:
  virtual ~MyAbstractClass();

  virtual void Method1();
  virtual void Method2();
  void Method3();

  virtual void Method4() = 0; // make MyAbstractClass not instantiable
};

In Windows programming, interfaces are fundamental in COM. In fact, a COM component exports only interfaces (i.e. pointers to v-tables , i.e. pointers to set of function pointers). This helps defining an ABI (Application Binary Interface) that makes it possible to e.g. build a COM component in C++ and use it in Visual Basic, or build a COM component in C and use it in C++, or build a COM component with Visual C++ version X and use it with Visual C++ version Y. In other words, with interfaces you have high decoupling between client code and server code.

Moreover, when you want to build DLL's with a C++ object-oriented interface (instead of pure C DLL's), as described in this article, it's better to export interfaces (the "mature approach") instead of C++ classes (this is basically what COM does, but without the burden of COM infrastructure).

I'd use an interface if I want to define a set of rules using which a component can be programmed, without specifying a concrete particular behavior. Classes that implement this interface will provide some concrete behavior themselves.

Instead, I'd use an abstract class when I want to provide some default infrastructure code and behavior, and make it possible to client code to derive from this abstract class, overriding the pure virtual methods with some custom code, and complete this behavior with custom code. Think for example of an infrastructure for an OpenGL application. You can define an abstract class that initializes OpenGL, sets up the window environment, etc. and then you can derive from this class and implement custom code for e.g. the rendering process and handling user input:

// Abstract class for an OpenGL app.
// Creates rendering window, initializes OpenGL; 
// client code must derive from it 
// and implement rendering and user input.
class OpenGLApp
{
public:
  OpenGLApp();
  virtual ~OpenGLApp();
  ...

  // Run the app    
  void Run();


  // <---- This behavior must be implemented by the client ---->

  // Rendering
  virtual void Render() = 0;

  // Handle user input
  // (returns false to quit, true to continue looping)
  virtual bool HandleInput() = 0;

  // <--------------------------------------------------------->


private:
  //
  // Some infrastructure code
  //
  ... 
  void CreateRenderingWindow();
  void CreateOpenGLContext();
  void SwapBuffers();
};


class MyOpenGLDemo : public OpenGLApp
{
public:
  MyOpenGLDemo();
  virtual ~MyOpenGLDemo();

  // Rendering
  virtual void Render();  // implements rendering code

  // Handle user input
  virtual bool HandleInput(); // implements user input handling


  //  ... some other stuff
};