Why is CGA color #6 brown instead of dark yellow?

Pure RGBI math makes color 6 dark yellow (#AAAA00): red and green on, blue off, no intensity. IBM considered that ugly and built a circuit into the 5153 monitor that halves the green signal for that one color, producing brown (#AA5500). Clone monitors without the circuit showed dark yellow.

How are CGA composite artifact colors generated?

On the composite output, the CGA pixel clock (14.31818 MHz) is exactly four times the NTSC color subcarrier (3.579545 MHz). A horizontal pattern of pixels therefore looks to a TV like a color subcarrier of a particular phase and amplitude, which the TV decodes as hue and saturation. The colors come from the dot pattern, not from the RGBI value of any single pixel.

How CGA color works

The IBM Color Graphics Adapter (1981) is simple hardware with a surprising amount of subtlety — and it's the subtlety that nearly every "retro palette" reference gets wrong. Here's what actually happens, from the 4-bit digital signal to the analog composite artifacts.

RGBI: four bits, sixteen colors

CGA's digital output has exactly four wires: R, G, B, and I (intensity). Four bits give 16 combinations — the 16 CGA colors. The mapping to actual brightness is straightforward: a primary that's "on" without intensity sits at about two-thirds level (0xAA = 170), and the intensity bit lifts the whole color by roughly one-third (0x55 = 85). So an "off" channel becomes 85 when intensity is set, and an "on" channel jumps to 255.

That's why dark gray is #555555 (intensity only, all primaries off → 85,85,85) and light blue is #5555FF (intensity + blue → 85,85,255). Every one of the 16 falls out of this rule — except one.

The brown exception. Color 6 is red + green, no intensity. By the rule that's #AAAA00 — dark yellow. IBM thought that looked bad, so the 5153 monitor contains a circuit that halves the green component for this one color, turning it into brown #AA5500. Clone monitors without that circuit showed dark yellow, and the retro community still argues about it. See the CGA palette page, which lets you toggle between the two.

Graphics modes: you don't get a free palette

In text mode all 16 colors are available. In graphics mode they are not freely choosable — and this is the second thing references get wrong.

320×200, 4 colors: each pixel is 2 bits. One of the four entries (the background) is freely chosen from all 16 RGBI colors. The other three come as a fixed group selected by two hardware bits — a palette-select bit (palette 0 = green/red/brown, palette 1 = cyan/magenta/white) and an intensity bit (dim vs. bright versions). There's also an officially undocumented mode 5 that exposes a third group (cyan/red/white). You can explore every combination in the CGA mode simulator.

640×200, 2 colors: one bit per pixel — a background and a single foreground. Higher resolution, less color.

Two outputs, two very different pictures

The CGA card has two physical outputs, and they don't show the same thing:

RGBI (the 9-pin digital connector) drives an RGB monitor like the 5153. Sharp pixels, the 16 colors above, no surprises.
Composite (the RCA jack) emits a single analog NTSC signal for a TV — and on a TV, the colors change completely.

Composite artifact colors: the real magic

Here's the key fact: CGA's pixel clock is 14.31818 MHz, which is exactly four times the NTSC color subcarrier of 3.579545 MHz. That 4:1 ratio is not a coincidence of convenience — it means a horizontal run of on/off pixels, emitted at the pixel clock, is electrically indistinguishable to a TV from a chroma subcarrier. The TV interprets it as color.

Specifically, within each group of four pixels (one subcarrier cycle): the phase of the pattern becomes hue, the amplitude becomes saturation, and the average level becomes brightness. A repeating dot pattern therefore is a color — one that may not exist among the 16 RGBI colors at all. In 640-pixel mode there are 16 distinct four-pixel patterns, yielding the classic ~16 artifact colors (the muddy browns, the artifact orange and blue, the grey you get from alternating pixels).

The universal bug. Almost every "composite CGA" tool fakes this by sampling the RGBI colors directly. That's wrong — it can never produce the artifact hues, because those colors come from the dot pattern across neighboring pixels, not from any single pixel's RGBI value. Doing it correctly means building the composite signal and demodulating it the way a TV does. Our image converter does exactly that (switch the monitor control to "NTSC composite"), and the method is documented on the accuracy page.

So which CGA colors are "right"?

All of them — it depends on the output and the monitor. The 16 RGBI colors are right for an RGB monitor (with the brown/dark-yellow split depending on the 5153 circuit). The artifact colors are right for a composite TV. A good tool models the hardware path rather than picking one answer and calling the others wrong. That's the whole idea behind this site.

Frequently asked

How many colors can CGA show at once?: 16 in text mode; 4 in 320×200 graphics (background + a fixed 3-color group); 2 in 640×200. On composite, dot patterns produce additional artifact colors beyond the 16.
Is composite color "real" color or a trick?: It's real in the sense that a TV genuinely displays it — but it's an artifact of how NTSC encodes color, not a palette the programmer selected directly. Developers learned to exploit it deliberately.
Where can I see the exact values?: The CGA, EGA, and VGA reference pages list every hex code; the simulator and converter show them live.

How this site models the hardware, and our sources →