When to use in vs ref vs out

You should use out unless you need ref.

It makes a big difference when the data needs to be marshalled e.g. to another process, which can be costly. So you want to avoid marshalling the initial value when the method doesn't make use of it.

Beyond that, it also shows the reader of the declaration or the call whether the initial value is relevant (and potentially preserved), or thrown away.

As a minor difference, an out parameter needs not be initialized.

Example for out:

string a, b;
person.GetBothNames(out a, out b);

where GetBothNames is a method to retrieve two values atomically, the method won't change behavior whatever a and b are. If the call goes to a server in Hawaii, copying the initial values from here to Hawaii is a waste of bandwidth. A similar snippet using ref:

string a = String.Empty, b = String.Empty;
person.GetBothNames(ref a, ref b);

could confuse readers, because it looks like the initial values of a and b are relevant (though the method name would indicate they are not).

Example for ref:

string name = textbox.Text;
bool didModify = validator.SuggestValidName(ref name);

Here the initial value is relevant to the method.

Use out to denote that the parameter is not being used, only set. This helps the caller understand that you're always initializing the parameter.

Also, ref and out are not just for value types. They also let you reset the object that a reference type is referencing from within a method.

You're correct in that, semantically, ref provides both "in" and "out" functionality, whereas out only provides "out" functionality. There are some things to consider:

1. out requires that the method accepting the parameter MUST, at some point before returning, assign a value to the variable. You find this pattern in some of the key/value data storage classes like Dictionary<K,V>, where you have functions like TryGetValue. This function takes an out parameter that holds what the value will be if retrieved. It wouldn't make sense for the caller to pass a value into this function, so out is used to guarantee that some value will be in the variable after the call, even if it isn't "real" data (in the case of TryGetValue where the key isn't present).
2. out and ref parameters are marshaled differently when dealing with interop code

Also, as an aside, it's important to note that while reference types and value types differ in the nature of their value, every variable in your application points to a location of memory that holds a value, even for reference types. It just happens that, with reference types, the value contained in that location of memory is another memory location. When you pass values to a function (or do any other variable assignment), the value of that variable is copied into the other variable. For value types, that means that the entire content of the type is copied. For reference types, that means that the memory location is copied. Either way, it does create a copy of the data contained in the variable. The only real relevance that this holds deals with assignment semantics; when assigning a variable or passing by value (the default), when a new assignment is made to the original (or new) variable, it does not affect the other variable. In the case of reference types, yes, changes made to the instance are available on both sides, but that's because the actual variable is just a pointer to another memory location; the content of the variable--the memory location--didn't actually change.

Passing with the ref keyword says that both the original variable and the function parameter will actually point to the same memory location. This, again, affects only assignment semantics. If a new value is assigned to one of the variables, then because the other points to the same memory location the new value will be reflected on the other side.

It depends on the compile context (See Example below).

out and ref both denote variable passing by reference, yet ref requires the variable to be initialized before being passed, which can be an important difference in the context of Marshaling (Interop: UmanagedToManagedTransition or vice versa)

MSDN warns:

Do not confuse the concept of passing by reference with the concept of reference types. The two concepts are not the same. A method parameter can be modified by ref regardless of whether it is a value type or a reference type. There is no boxing of a value type when it is passed by reference.

From the official MSDN Docs:

• out:

The out keyword causes arguments to be passed by reference. This is similar to the ref keyword, except that ref requires that the variable be initialized before being passed

• ref:

The ref keyword causes an argument to be passed by reference, not by value. The effect of passing by reference is that any change to the parameter in the method is reflected in the underlying argument variable in the calling method. The value of a reference parameter is always the same as the value of the underlying argument variable.

We can verify that the out and ref are indeed the same when the argument gets assigned:

CIL Example:

Consider the following example

static class outRefTest{
    public static int myfunc(int x){x=0; return x; }
    public static void myfuncOut(out int x){x=0;}
    public static void myfuncRef(ref int x){x=0;}
    public static void myfuncRefEmpty(ref int x){}
    // Define other methods and classes here
}

in CIL, the instructions of myfuncOut and myfuncRef are identical as expected.

outRefTest.myfunc:
IL_0000:  nop         
IL_0001:  ldc.i4.0    
IL_0002:  starg.s     00 
IL_0004:  ldarg.0     
IL_0005:  stloc.0     
IL_0006:  br.s        IL_0008
IL_0008:  ldloc.0     
IL_0009:  ret         

outRefTest.myfuncOut:
IL_0000:  nop         
IL_0001:  ldarg.0     
IL_0002:  ldc.i4.0    
IL_0003:  stind.i4    
IL_0004:  ret         

outRefTest.myfuncRef:
IL_0000:  nop         
IL_0001:  ldarg.0     
IL_0002:  ldc.i4.0    
IL_0003:  stind.i4    
IL_0004:  ret         

outRefTest.myfuncRefEmpty:
IL_0000:  nop         
IL_0001:  ret         

nop: no operation, ldloc: load local, stloc: stack local, ldarg: load argument, bs.s: branch to target....

(See: List of CIL instructions )

Below are some notes which i pulled from this codeproject article on C# Out Vs Ref

1. It should be used only when we are expecting multiple outputs from a function or a method. A thought on structures can be also a good option for the same.
2. REF and OUT are keywords which dictate how data is passed from caller to callee and vice versa.
3. In REF data passes two way. From caller to callee and vice-versa.
4. In Out data passes only one way from callee to caller. In this case if Caller tried to send data to the callee it will be overlooked / rejected.

If you are a visual person then please see this yourtube video which demonstrates the difference practically https://www.youtube.com/watch?v=lYdcY5zulXA

Below image shows the differences more visually