Segment overrides

Each memory operand has a default segment: SS when the address uses BP, DS otherwise. Writing a segment register and a colon before the operand, as in MOV AX, ES:[BX], makes the CPU use that segment instead. The assembler turns it into a one-byte prefix in front of the instruction. It applies to that instruction only, and changes no flags.

MOV     AX, 0B800h
MOV     ES, AX       ; ES -> colour text memory
MOV     DI, 140      ; row 0, column 70 (two bytes per cell)
MOV     SI, OFFSET TXT1
MOV     CX, 5
COPY1:
MOV     AL, CS:[SI]  ; read our data through CS
MOV     AH, 1Eh      ; yellow on blue
MOV     ES:[DI], AX  ; write through ES
ADD     DI, 2
INC     SI
LOOP    COPY1
 
TXT1    DB      'HELLO'

What cannot be overridden

Instruction fetch always uses CS:IP. The stack operations (PUSH, POP, CALL, RET, interrupts) always use SS:SP. The destination of a string instruction is always ES:DI. The source of a string instruction, DS:SI, can be overridden, so MOVS can copy from ES or CS to ES.

Why an interrupt handler uses CS:

When a hardware interrupt arrives, DS still belongs to whichever program was running. A handler that keeps its variables in its own code segment reads them with CS: until it has set DS, or without ever touching DS.

Pitfalls

In a .COM all segment registers start equal, so a missing override can go unnoticed until the code moves into a .EXE. In a .EXE, SS and DS usually differ, and [BP+...] pointing at data needs DS:. MASM also inserts overrides on its own, based on ASSUME, when a variable lives in a segment that DS is not assumed to hold. On the 8086 and 8088, a repeated string instruction with both REP and a segment prefix can lose one of its prefixes if an interrupt arrives in the middle; disable interrupts around it or avoid the combination.

Later CPUs

The 80386 adds FS: (64h) and GS: (65h).