What is most common and correct practice to get a CGFloat from an NSNumber?

This will get the correct result in any case:

 NSNumber *n = @42.42;
 CGFloat cgf = [n doubleValue];

because CGFloat is either float or double.

NSNumber does not have a CGFloatValue method. You could define one using the toll-free bridge to CFNumberRef:

@interface NSNumber (MyCGFloatValue)
-(CGFloat)myCGFloatValue;
@end

@implementation NSNumber (MyCGFloatValue)
-(CGFloat)myCGFloatValue{
    CGFloat result;
    CFNumberGetValue((__bridge CFNumberRef)(self), kCFNumberCGFloatType, &result);
    return result;
}
@end

or using the C11 feature "Generic selection", where the compiler chooses the appropriate code depending on the type of CGFloat:

@implementation NSNumber (MyCGFloatValue)
-(CGFloat)myCGFloatValue{
    CGFloat result;
    result = _Generic(result,
            double: [self doubleValue],
            float: [self floatValue]);
    return result;
}
@end

And then

NSNumber *n = @42.24;
CGFloat f = [n myCGFloatValue];

but I doubt that it is worth the hassle.

Just always get the double value. If CGFloat is only a ‘float’ on your machine the double will be instantly truncated, so you don’t care.

The older answers are both correct. Just for my 2¢, here's my implementation:

@implementation NSNumber (TypeConversion)

- (CGFloat)cgFloatValue
{
#if CGFLOAT_IS_DOUBLE
    return self.doubleValue;
#else
    return self.floatValue;
#endif
}

@end

This relies on the definition of CGFLOAT_IS_DOUBLE in CGBase.h for Objective C, or CoreGraphics.h for Swift:

// Don't write this code; it's already included in Foundation :)

#if defined(__LP64__) && __LP64__
# define CGFLOAT_TYPE double
# define CGFLOAT_IS_DOUBLE 1
# define CGFLOAT_MIN DBL_MIN
# define CGFLOAT_MAX DBL_MAX
#else
# define CGFLOAT_TYPE float
# define CGFLOAT_IS_DOUBLE 0
# define CGFLOAT_MIN FLT_MIN
# define CGFLOAT_MAX FLT_MAX
#endif

typedef CGFLOAT_TYPE CGFloat;
#define CGFLOAT_DEFINED 1

^{From CoreGraphics.h}