Resistor Divider Calculator

Enter an input, an output and a divider current. The calculator picks real E-series values, then reports the worst-case output range with the resistor tolerance and the input supply tolerance stacked together — plus dissipation and a package suggestion, and which of the three error sources is actually hurting you.

Runs entirely in your browser. Nothing is uploaded, stored, or sent anywhere.

Why a divider is rarely as accurate as it looks

The ratio is exact arithmetic, so a divider feels like a precise thing. It is not. Three separate errors stack on the output, and only one of them is the resistors.

The first is the supply. A divider divides whatever it is given, so every percent of error on V_in appears as a percent on V_out, untouched. If you are dividing a rail that is itself ±2%, no choice of resistor makes the output better than ±2%. This is the error most often left out of the calculation, and it is why the tool asks for it.

The second is resistor tolerance — and it is smaller than people expect, because the two errors partially cancel. Both resistors appear in the denominator, so if both drift the same way the ratio barely moves. The worst case is when they drift in opposite directions, which is what this tool computes.

How much cancellation you get depends on the ratio, and the factor is exactly 1 − Vout/Vin. A 2:1 divider gets the most: two ±1% parts give about ±1% out, not ±2%. A deep step-down gets almost none — dividing 12 V to 3.3 V, the factor is 0.73, so the same two parts give about ±1.5%. It is worth knowing which end of that you are on before assuming the resistors are the cheap part of your error budget.

The third is E-series rounding, and it is not a tolerance at all — it is a fixed offset. The nearest stock value is not your ideal value, so the nominal output sits slightly off target before any part varies. It does not average out across a production run; every board has it, in the same direction. The tool reports it separately for that reason.

What the E-series actually is

Stock resistor values are not arbitrary. Each series divides a decade into logarithmically even steps: E24 gives 24 values per decade, E96 gives 96. The spacing is chosen so that consecutive values are about one tolerance band apart, which is why the series and the tolerance travel together — E24 with ±5%, E96 with ±1%. Selecting a series here sets the tolerance to its conventional partner, and you can override it: buying ±1% parts in E24 values is perfectly normal, and it buys accuracy without a wider value choice.

Going to a finer series only helps if the rounding error is what is hurting you. If the supply dominates, E192 changes almost nothing — the verdict above says which case you are in rather than leaving you to guess.

How the pairs are chosen

Rather than picking a resistance and rounding it, the tool searches the whole series. For every stock value of R2 the ideal R1 is exact, so only the two series values bracketing it can possibly win — that makes an exhaustive search cheap. Each candidate is then filtered against your two real constraints, the power rating after derating and the current cap, and what survives is ranked by ratio error.

Any ratio recurs in every decade, so the table lists each ratio once, at its lowest-resistance decade. That is deliberate: within the current cap you generally want the lowest resistance you can afford. High-value dividers pick up noise, are walked over by leakage and input bias, load more easily, and settle slowly into a sampling ADC. The current cap is what stops the search going too far the other way — on a battery product a 1 mA divider is often the largest standing load on the board.

A finer series is not automatically a better fit

It is natural to assume E96 must beat E24, and for most ratios it does. But E24 is not a subset of E96 — they are independently defined series, and E24 contains values such as 24 and 91 that have no exact counterpart in E96. So a particular ratio can land almost perfectly on a cheap E24 pair and awkwardly between E96 values.

The defaults show it. Dividing 12 V to 3.3 V, E24 offers 24 kΩ over 9.1 kΩ for a ratio error of about −0.03%. The best E96 pair is roughly +0.47% — more than ten times worse, with tighter and more expensive parts. Change the series in the tool and watch the top of the table. It is always worth checking both before assuming the finer series is the answer.

Note too what a finer series does and does not buy. It reduces the rounding error, which is a fixed offset. It does nothing to the supply tolerance, and it only helps the resistor tolerance if you actually buy the tighter parts — the series and the tolerance are separate choices, which is why you can override the tolerance field independently.

Dissipation and derating

The headroom column is the ratio of your derated budget to the worst-case dissipation, so anything at or near 1× is running at the limit you set. The default derating of 60% is a common working figure, not a rule: a 0603 rated 0.1 W at 70 °C is not rated 0.1 W at 105 °C, and every manufacturer publishes a derating curve that slopes to zero at the part’s maximum temperature. Running near the limit is also how resistors drift out of tolerance over life, which quietly undoes the accuracy you paid for.

Loading is the error that surprises people

A divider is a source with an output impedance of R1 ∥ R2. Anything you connect across the output loads it, and loading only ever pulls the output down. Enter a load resistance and the tool folds it into the nominal, the worst case and the contribution table.

The rule of thumb is that the load should be at least ten times the divider’s output impedance for the error to stay under about 10%, and a hundred times for it to be negligible. If you are driving an ADC input, remember that the sampling capacitor is a dynamic load and a high-impedance divider may not settle inside the acquisition window — a buffer, not a smaller resistor, is usually the right fix.

Noise, and the part that depends on resistor type

A divider has two noise mechanisms, and only one of them cares what the resistor is made of. Conflating them is why “use better resistors” is sometimes transformative and sometimes does nothing at all.

Thermal noise is set by the source impedance, the temperature and the bandwidth — nothing else. A thick-film and a metal-foil resistor of the same value produce identical thermal noise. It is a floor you cannot buy your way under; you can only lower it by reducing R1 ∥ R2 or narrowing the bandwidth, and it falls only as the square root of either.

Excess noise, also called 1/f or current noise, is the type-dependent one, and it exists only because DC current is flowing — which is a divider’s permanent condition. It is quantified by noise index: microvolts of noise per volt of DC across the part, per decade of frequency. The spread across constructions is enormous, roughly 2 µV/V for carbon composition, 0.1 for thick film, 0.03 for thin film, and under 0.01 for metal foil and wirewound. That is a factor of a few hundred between the cheapest and the best part on an otherwise identical schematic.

Both resistors contribute, and each reaches the tap through the divider’s own transfer function. The two terms turn out exactly equal — R1’s noise is attenuated by the ratio while R2’s is attenuated by its complement, and the products match — which is why the expression above collapses to a single term with a √2 on it.

Which one dominates flips with bandwidth

This is the useful part. Thermal noise grows with the square root of bandwidth; excess noise grows with the square root of decades. Those scale completely differently, so the answer to “does resistor type matter here” depends entirely on how wide your measurement is.

Take the default divider. Over 1 Hz to 100 kHz, thermal noise is about 3.3 µV and thick-film excess noise about 0.76 µV — thermal wins, and switching to metal foil changes the total by 3%. Now narrow to 1–10 Hz, the bandwidth that actually matters for a voltage reference, a slow ADC or a comparator threshold: thermal collapses to 0.03 µV while excess noise only falls to 0.34 µV. Excess noise now dominates by more than ten to one, and changing resistor type buys an order of magnitude that no tolerance grade could. The tool says which regime you are in rather than leaving you to work it out.

One related type effect is not noise and is not modelled here: thick film has a voltage coefficient, so its resistance shifts slightly with the voltage across it. In a divider the two halves see different voltages, so it does not cancel — it shows up as a small non-linearity if the input swings.

What this does not cover

• Temperature coefficient. Thick-film parts are commonly 100–200 ppm/°C. Over a 100 °C swing that is another 1–2%, and it does not cancel between the two resistors unless they are the same type, the same size and at the same temperature. A matched divider network exists precisely for this.
• Drift and humidity over life, which for thick film can equal the initial tolerance again after years in the field.
• Self-heating shifting a part’s own value once dissipation is significant.
• Input bias and leakage at the tap, which behave like an extra load and matter most in exactly the high-impedance dividers chosen to save power.
• Worst case is computed as a hard corner — every tolerance simultaneously at its extreme. That is correct for a guaranteed-by-design argument. If you want a realistic production distribution instead, an RSS or Monte Carlo treatment gives a much tighter and equally defensible number for non-safety work.

Where this fits

Board-level analog and digital design is our primary service line — feedback dividers, references, analog front-ends and the layout around them. If a divider is setting something that matters, a regulator’s feedback node or a comparator threshold, the tolerance stack is worth doing properly rather than assuming the nominal.

More free tools

Each of these runs entirely in your browser. Nothing is uploaded, stored or sent anywhere, and none of them needs an email address.

See all 26 engineering tools →