How to implement a read only property

C# 6.0 adds readonly auto properties

public object MyProperty { get; }

So when you don't need to support older compilers you can have a truly readonly property with code that's just as concise as a readonly field.

---

Versioning:
I think it doesn't make much difference if you are only interested in source compatibility.
Using a property is better for binary compatibility since you can replace it by a property which has a setter without breaking compiled code depending on your library.

Convention:
You are following the convention. In cases like this where the differences between the two possibilities are relatively minor following the convention is better. One case where it might come back to bite you is reflection based code. It might only accept properties and not fields, for example a property editor/viewer.

Serialization
Changing from field to property will probably break a lot of serializers. And AFAIK XmlSerializer does only serialize public properties and not public fields.

Using an Autoproperty
Another common Variation is using an autoproperty with a private setter. While this is short and a property it doesn't enforce the readonlyness. So I prefer the other ones.

Readonly field is selfdocumenting
There is one advantage of the field though:
It makes it clear at a glance at the public interface that it's actually immutable (barring reflection). Whereas in case of a property you can only see that you cannot change it, so you'd have to refer to the documentation or implementation.

But to be honest I use the first one quite often in application code since I'm lazy. In libraries I'm typically more thorough and follow the convention.

The second way is the preferred option.

private readonly int MyVal = 5;

public int MyProp { get { return MyVal;}  }

This will ensure that MyVal can only be assigned at initialization (it can also be set in a constructor).

As you had noted - this way you are not exposing an internal member, allowing you to change the internal implementation in the future.

With the introduction of C# 6 (in VS 2015), you can now have get-only automatic properties, in which the implicit backing field is readonly (i.e. values can be assigned in the constructor but not elsewhere):

public string Name { get; }

public Customer(string name)  // Constructor
{
    Name = name;
}

private void SomeFunction()
{
    Name = "Something Else";  // Compile-time error
}

And you can now also initialise properties (with or without a setter) inline:

public string Name { get; } = "Boris";

Referring back to the question, this gives you the advantages of option 2 (public member is a property, not a field) with the brevity of option 1.

Unfortunately, it doesn't provide a guarantee of immutability at the level of the public interface (as in @CodesInChaos's point about self-documentation), because to a consumer of the class, having no setter is indistinguishable from having a private setter.